Professional Documents
Culture Documents
Lehmanexoticcredderivs PDF
Lehmanexoticcredderivs PDF
.. Portfolio Swaps
Credit Index Products
.. Repackagings
Default Baskets
. Default swaptions
Credit hybrids
Leadership in Fixed Income
Research For further information please contact
your local sales representative or call:
London: Giancarlo Saronne
+44 20 7260 2745 gsaronne@lehman.com
All Rights Reserved. Member SIPC. Lehman Brothers International (Europe) is regulated by the Financial Services Authority. 2003 Lehman Brothers Inc.
guide.qxd 10/10/2003 11:15 Page 1
Foreword
The credit derivatives market has revolu- thetic loss tranches and default baskets cre-
tionised the transfer of credit risk. Its impact ate new risk-return profiles to appeal to the
has been borne out by its significant growth differing risk appetites of investors based on
which has currently achieved a market notion- the tranching of portfolio credit risk. In doing
al close to $2 trillion. While not directly com- so they create an exposure to default correla-
parable, it is worth noting that the total tion. CDS options allow investors to express
notional outstanding of global investment a view on credit spread volatility, and hybrid
grade corporate bond issuance currently products allow investors to mix credit risk
stands at $3.1 trillion. views with interest rate and FX risk.
This growth in the credit derivatives market More recently, we have seen a stepped
has been driven by an increasing realisation increase in the liquidity of these exotic credit
of the advantages credit derivatives possess derivative products. This includes the devel-
over the cash alternative, plus the many new opment of very liquid portfolio credit vehicles,
possibilities they present to both credit the arrival of a two-way correlation market in
investors and hedgers. Those investors seek- customised CDO tranches, and the develop-
ing diversification, yield pickup or new ways ment of a more liquid default swaptions mar-
to take an exposure to credit are increasingly ket. To enable this growth, the market has
turning towards the credit derivatives market. developed new approaches to the pricing and
The primary purpose of credit derivatives is risk-management of these products.
to enable the efficient transfer and repack- As a result, this book is divided into two
aging of credit risk. In their simplest form, parts. In the first half, we describe how exotic
credit derivatives provide a more efficient structured credit products work, their ratio-
way to replicate in a derivative format the nale, risks and uses. In the second half, we
credit risks that would otherwise exist in a review the models for pricing and risk manag-
standard cash instrument. ing these various credit derivatives, focusing
More exotic credit derivatives such as syn- on implementation and calibration issues.
Authors
Dominic O'Kane Claus Pedersen
T. +44 207 260 2628 T. +1 212 526 7775
E. dokane@lehman.com E. cmpeders@lehman.com
Marco Naldi Lutz Schloegl
T. +1 212 526 1728 T. +44 207 260 2113
E. mnaldi@lehman.com E. luschloe@lehman.com
Sunita Ganapati Roy Mashal
T. +1 415 274 5485 T. +1 212 526 7931
E. sganapati@lehman.com E. rmashal@lehman.com
Arthur Berd
T. +1 212 526 2629
E. arthur.berd@lehman.com
Contents
Foreword 1
References 53
bankruptcy, failure to pay and restructuring. amount of deliverable obligations of the ref-
Buying credit protection is economically erence entity to the protection seller in
equivalent to shorting the credit risk. Equally, return for the notional amount paid in cash.
selling credit protection is economically In general there are several deliverable obli-
equivalent to going long the credit risk. gations from which the protection buyer can
This protection lasts until some specified choose which satisfy a number of character-
maturity date. For this protection, the pro- istics. Typically they include restrictions on
tection buyer makes quarterly payments, to the maturity and the requirement that they
the protection seller, as shown in Figure 3, be pari passu most CDS are linked to
until a credit event or maturity, whichever senior unsecured debt.
occurs first. This is known as the premium If the deliverable obligations trade with dif-
leg. The actual payment amounts on the pre- ferent prices following a credit event, which
mium leg are determined by the CDS spread they are most likely to do if the credit event
adjusted for the frequency using a basis is a restructuring, the protection buyer can
convention, usually Actual 360. take advantage of this situation by buying
If a credit event does occur before the and delivering the cheapest asset. The pro-
maturity date of the contract, there is a pay- tection buyer is therefore long a cheapest to
ment by the protection seller to the protec- deliver option.
tion buyer. We call this leg of the CDS the
protection leg. This payment equals the dif- Cash settlement This is the alternative to
ference between par and the price of the physical settlement, and is used less fre-
assets of the reference entity on the face quently in standard CDS but overwhelming-
value of the protection, and compensates the ly in tranched CDOs, as discussed later. In
protection buyer for the loss. There are two cash settlement, a cash payment is made by
ways to settle the payment of the protection the protection seller to the protection buyer
leg, the choice being made at the initiation of equal to par minus the recovery rate of the
the contract. They are: reference asset. The recovery rate is calcu-
lated by referencing dealer quotes or
Physical settlement This is the most wide- observable market prices over some period
ly used settlement procedure. It requires the after default has occurred.
protection buyer to deliver the notional Suppose a protection buyer purchases
five-year protection on a company at a CDS
spread of 300bp. The face value of the pro-
Figure 3. Mechanics of a CDS tection is $10m. The protection buyer
therefore makes quarterly payments ap-
proximately (we ignore calendars and day
Default swap spread count conventions) equal to $10m 0.03
Protection (premium leg)
Protection
buyer seller
0.25 = $75,000. After a short period the
reference entity suffers a credit event.
Contingent payment of loss on par Assuming that the cheapest deliverable
following a credit event (protection leg) asset of the reference entity has a recovery
price of $45 per $100 of face value, the pay-
ments are as follows:
The protection seller compensates the This can be done for long periods without
protection buyer for the loss on the face assuming any repo risk. This is very use-
value of the asset received by the protec- ful for those wishing to hedge current
tion buyer and this is equal to $5.5m. credit exposures or those wishing to take
a bearish credit view.
The protection buyer pays the accrued
premium from the previous premium CDS are unfunded so leverage is possi-
payment date to the time of the credit ble. This is also an advantage for those
event. For example, if the credit event who have high funding costs, because
occurs after a month then the protection CDS implicitly lock in Libor funding to
buyer pays approximately $10m 300bp maturity.
1/12 = $25,000 of premium accrued.
Note that this is the standard for corpo- CDS are customisable, although devia-
rate reference entity linked CDS. tion from the standard may incur a liquid-
ity cost.
For severely distressed reference entities,
the CDS contract trades in an up-front for- CDS can be used to take a spread view
mat where the protection buyer makes a on a credit, as with a bond.
cash payment at trade initiation which pur-
chases protection to some specified maturi- Dislocations between cash and CDS pre-
ty there are no subsequent payments sent new relative value opportunities.
unless there is a credit event in which the This is known as trading the default
protection leg is settled as in a standard swap basis.
CDS. For a full description of up-front CDS
see OKane and Sen (2003). Evolution of CDS documentation
Liquidity in the CDS market differs from The CDS is a contract traded within the legal
the cash credit market in a number of ways. framework of the International Swaps and
For a start, a wider range of credits trade in Derivatives Association (ISDA) master agree-
the CDS market than in cash. In terms of ment. The definitions used by the market for
maturity, the most liquid CDS is the five-year credit events and other contractual details
contract, followed by the three-year, seven- have been set out in the ISDA 1999 document
year and 10-year. The fact that a physical and recently amended and enhanced by the
asset does not need to be sourced means ISDA 2003 document. The advantage of this
that it is generally easier to transact in large standardisation of a unique set of definitions
round sizes with CDS. is that it reduces legal risk, speeds up the con-
firmation process and so enhances liquidity.
Uses of a CDS Despite this standardisation of defini-
The CDS can do almost everything that cash tions, the CDS market does not have a uni-
can do and more. We list some of the main versal standard contract. Instead, there is a
applications of CDS below. US, European and an Asian market stan-
dard, differentiated by the way they treat a
The CDS has revolutionised the credit restructuring credit event. This is the con-
markets by making it easy to short credit. sequence of a desire to enhance the posi-
tion of protection sellers by limiting the For that we need a model and a discussion of
value of the protection buyers delivery the valuation of CDS is provided on page 32.
option following a restructuring credit
event. A full discussion and analysis of Funded versus unfunded
these different standards can be found in Credit derivatives, including CDS, can be
OKane, Pedersen and Turnbull (2003). traded in a number of formats. The most
commonly used is known as swap format,
Determining the CDS spread and this is the standard for CDS. This format
The premium payments in a CDS are is also termed unfunded format because
defined in terms of a CDS spread, paid peri- the investor makes no upfront payment.
odically on the protected notional until Subsequent payments are simply payments
maturity or a credit event. It is possible to of spread and there is no principal payment
show that the CDS spread can, to a first at maturity. Losses require payments to
approximation, be proxied by either (i) a par be made by the protection seller to the
floater bond spread (the spread to Libor at protection buyer, and this has counterparty
which the reference entity can issue a float- risk implications.
ing rate note of the same maturity at a price The other format is to trade the risk in the
of par) or (ii) the asset swap spread of a form of a credit linked note. This format is
bond of the same maturity provided it known as funded because the investor has
trades close to par. to fund an initial payment, typically par. This
Demonstrating these relationships relies par is used by the protection buyer to pur-
on several assumptions that break down in chase high quality collateral. In return the pro-
practice. For example, we assume a com- tection seller receives a coupon, which may
mon market-wide funding level of Libor, we be floating rate, ie, Libor plus a spread, or
ignore accrued coupons on default, we may be fixed at a rate above the same matu-
ignore the delivery option in the CDS, and rity swap rate. At maturity, if no default has
we ignore counterparty risk. Despite these occurred the collateral matures and the
assumptions, cash market spreads usually investor is returned par. Any default before
provide the starting point for where CDS maturity results in the collateral being sold,
spreads should trade. The difference the protection buyer covering his loss and the
between where CDS spreads and cash investor receiving par minus the loss. The
LIBOR spreads trade is known as the protection buyer is exposed to the default risk
Default Swap Basis, defined as: of the collateral rather than the counterparty.
Basis = CDS Spread Cash Libor Spread. Traded CDS portfolio products
CDS portfolio products are products that
A full discussion of the drivers behind the enable the investor to go long or short the
CDS basis is provided in OKane and credit risk associated with a portfolio of CDS
McAdie (2001). A large number of in one transaction.
investors now exploit the basis as a rela- In recent months, we have seen the emer-
tive value play. gence of a number of very liquid portfolio
Determining the CDS spread is not the products, whose aim is to offer investors a
same as valuing an existing CDS contract. diverse, liquid vehicle for assuming or hedg-
ing exposure to different credit markets, one of a first-to-default (FTD) basket, n=1, and it
example being the TRAC-XSM vehicle. These is the first credit in a basket of reference
have added liquidity to the CDS market and credits whose default triggers a payment to
also created a standard which can be used the protection buyer. As with a CDS, the con-
to develop portfolio credit derivatives such tingent payment typically involves physical
as options on TRAC-X. delivery of the defaulted asset in return for a
The move of the CDS market from banks payment of the par amount in cash. In return
towards traditional credit investors has greatly for assuming the nth-to-default risk, the pro-
increased the need for a performance bench- tection seller receives a spread paid on the
mark linked directly to the CDS market. As a notional of the position as a series of regular
consequence, Lehman Brothers has intro- cash flows until maturity or the nth credit
duced a family of global investment grade CDS event, whichever is sooner.
indices which are discussed in Munves (2003). The advantage of an FTD basket is that it
These consist of three sub-indices, a US enables an investor to earn a higher yield
250 name index, a European 150 name index than any of the credits in the basket. This is
and a Japanese 40 name index. All names because the seller of FTD protection is lever-
are corporates and the maturity of the index aging their credit risk.
is maintained close to five years. Daily pric- To see this, consider that the fair-value
ing of all 440 names is available on our spread paid by a credit risky asset is deter-
LehmanLive website. mined by the probability of a default, times
the size of the loss given default. FTD bas-
Basket default swaps kets leverage the credit risk by increasing the
Correlation products are based on redistribut- probability of a loss by conditioning the pay-
ing the credit risk of a portfolio of single- off on the first default among several credits.
name credits across a number of different The size of the potential loss does not
securities. The portfolio may be as small as increase relative to buying any of the assets
five credits or as large as 200 or more credits. in the basket. The most that the investor can
The redistribution mechanism is based on the lose is par minus the recovery value of the
idea of assigning losses on the credit portfo- FTD asset on the face value of the basket.
lio to the different securities in a specified pri- The advantage is that the basket spread
ority, with some securities taking the first paid can be a multiple of the spread paid by
losses and others taking later losses. This the individual assets in the basket. This is
exposes the investor to the tendency of shown in Figure 4 where we have a basket
assets in the portfolio to default together, ie, of five investment grade credits paying an
default correlation. The simplest correlation average spread of about 28bp. The FTD bas-
product is the basket default swap. ket pays a spread of 120bp.
A basket default swap is similar to a CDS, More risk-averse investors can use default
the difference being that the trigger is the baskets to construct lower risk assets: sec-
nth credit event in a specified basket of ref- ond-to-default (STD) baskets, where n=2,
erence entities. Typical baskets contain five trigger a credit event after two or more assets
to 10 reference entities. In the particular case have defaulted. As such they are lower risk
second-loss exposure products which will
TRAC-X is a service mark of JPMorgan and Morgan Stanley pay a lower spread than an FTD basket.
60
defaulting. Because an FTD is triggered
50
by only one credit event, it will be as risky
40
30
as the riskiest asset and the FTD basket
20 spread should be the widest spread of
10 the credits in the basket.
0
0 1 2 3 4 5
Number of defaults
The best way to understand the behaviour
of default baskets between these two cor-
relation limits is to study the loss distribu-
that if two assets are correlated, they will tion for the basket portfolio. See page 33
not only tend to default together, they will for a discussion of how to model the loss
also tend to survive together. distribution.
There are two correlation limits in which a We consider a basket of five credits with
FTD basket can be priced without resorting spreads of 100bp and an assumed recovery
to a model independence and maximum rate for all of 40%. We have plotted the loss
correlation. distribution for correlations of 0%, 20%, and
50% in Figure 5. The spread for an FTD bas-
Independence: Consider a five-credit ket depends on the probability of one or
basket where all of the underlying credits more defaults which equals one minus the
have flat credit curves. If the credits are probability of no defaults. We see that the
all independent and never become corre- probability of no defaults increases with
lated during the life of the trade, the nat- increasing correlation the probability of
ural hedge is for the basket investor to credits surviving together increases and
buy CDS protection on each of the indi- the FTD spread should fall.
vidual names to the full notional. If a The risk of an STD basket depends on the
credit event occurs, the CDS hedge cov- probability of two or more defaults. As corre-
ers the loss on the basket and all of the lation goes up from 020%, the probability of
other CDS hedges can be unwound at no two, three, four and five defaults increases.
cost, since they should on average have This makes the STD spread increase.
rolled down their flat credit curves. This The process for translating these loss dis-
implies that the basket spread for tributions into a fair value spread requires a
independent assets should be equal to model of the type described on page 39.
the sum of the spreads of the names in Essentially we have to find the basket
the basket. spread for which the present value of the
protection payments equals the present
Maximum correlation: Consider the value of the premium payments.
same FTD basket but this time where the We should not forget that in addition to the
protection leg, the premium leg of the The basket can be customised to the
default basket also has correlation sensitivi- investors exact view regarding size,
ty because it is only paid for as long as the maturity, number of credits, credit selec-
nth default does not occur. tion, FTD or STD.
Using a model we have calculated the cor-
relation sensitivity of the FTD and STD spread Buy and hold investors can enjoy the
for the five-credit basket shown in Figure 6. leveraging of the spread premium. This is
At low correlation, the FTD spread is close to discussed in more detail later.
146bp, which is the sum of the spreads. At
high correlation, the basket has the risk of the Credit investors can use default baskets
widest spread asset and so is at 30bp. The to hedge a blow-up in a portfolio of cred-
STD spread is lowest at zero correlation since its more cheaply than buying protection
the probability of two assets defaulting is low on the individual credits.
if the assets are independent. At maximum
correlation the STD spread tends towards the Default baskets can be used to express a
spread of the second widest asset in the bas- view on default correlation. If the
ket which is also 30bp. investors view is that the implied correla-
tion is too low then the investor should sell
Applications FTD protection. If implied correlation is too
Baskets have a range of applications. high they should sell STD protection.
Investors can use default baskets to lever-
age their credit exposure and so earn a Hedging default baskets
higher yield without increasing their The issuers of default baskets need to
notional at risk. hedge their risks. Spread risk is hedged by
selling protection dynamically in the CDS
The reference entities in the basket are all market on all of the credits in the default
typically investment grade and so are basket. Determining how much to sell,
familiar to most credit analysts. known as the delta, requires a pricing model
to calculate the sensitivity of the basket
Figure 6. Correlation dependence of value to changes in the spread curve of the
spread for FTD and STD basket underlying credit.
Although this delta hedging should immu-
nise the dealers portfolio against small
140
changes in spreads, it is not guaranteed to be
Basket spread (bp)
120
FTD a full hedge against a sudden default. For
100
80 STD instance, a dealer hedging an FTD basket
60 where a credit defaults with a recovery rate of
40
R would receive a payment of (1-R)F from the
20
0 protection seller, and will pay D(1-R)F on the
0 10 20 30 40 50 60 70 80 90 100 hedged protection, where F is the basket face
Correlation (%)
value and D is the delta in terms of percent-
age of face value. The net payment to the
protection buyer is therefore (1-D)(1-R)F.
There will also probably be a loss on the Hedgers of long protection FTD baskets
other CDS hedges. The expected spread are also long gamma. This means that as the
widening on default on the other credits in spread of an asset widens, the delta will
the basket due to their positive correlation increase and so the hedger will be selling
with the defaulted asset will result in a loss protection at a wider spread. If the spread
when they are unwound. The greater unwind tightens, then the delta will fall and the
losses for baskets with higher correlations hedger will be buying back hedges at a
will be factored into the basket spread. tighter level. So spread volatility can be ben-
One way for a default basket dealer to eficial. This effect helps to offset the nega-
reduce his correlation risk is by selling pro- tive carry associated with hedged FTD
tection on the same or similar default bas- baskets. This is clear in the previous exam-
kets. However this is difficult as it is usually ple where the income from the hedges is
difficult to find protection buyers who select 211bp, lower than the 246bp paid to the FTD
the exact same basket as an investor. basket investor.
The alternative hedging approach is for the Different rating agencies have developed
dealer to buy protection using default bas- their own model-based approaches for the
kets on other orders of protection. This is rating of default baskets. We discuss these
based on the observation that a dealer who on page 39.
is long first, second, third up to Mth order
protection on an M-credit basket has almost Synthetic CDOs
no correlation risk, since this position is Synthetic collateralised debt obligations
almost economically equivalent to buying (Synthetic CDOs) were conceived in 1997 as
full face value protection using CDS on all M a flexible and low-cost mechanism for trans-
credits in the basket. ferring credit risk off bank balance sheets.
Figure 7 shows an example basket with The primary motivation was the banks
the delta and spread for each of the five reduction of regulatory capital.
credits. Note that the deltas are all very More recently, however, the fusion of cred-
similar. This reflects the fact that all of the it derivatives modelling techniques and
assets have a similar spread. Differences derivatives trading have led to the creation of
are mainly due to our different correlation a new type of synthetic CDO, which we call
assumptions. a customised CDO, which can be tailored to
the exact risk appetites of different classes
Figure 7. Default basket deltas for a of investors. As a result, the synthetic CDO
10m notional five-year FTD basket on has become an investor-driven product.
five credits. The FTD spread is 246bp. Overall, these different types of synthetic
Reference entity CDS Spread Delta CDO have a total market size estimated by
Walt Disney 62bp 6.26m the Risk 2003 survey to be close to $500 bil-
Rolls Royce 60bp 6.55m lion. What is also of interest is that the deal-
Sun Microsystems 60bp 6.87m er-hedging of these products in the CDS
market has generated a substantial demand
Eastman Chemical 60bp 7.16m
to sell protection, balancing the traditional
France Telecom 64bp 7.57m
protection-buying demand coming from
bank loan book managers.
Subordinated
Super swap
senior swap premium Issued Senior notes
premium Special notes AAA
Highly Sponsoring purpose
rated bank vehicle Mezzanine notes
counterparty (SPV) BBB/A
$900m 10% first
loss Proceeds
super subordinated Equity notes (unrated)
senior credit
credit protection
protection
Credit CDS spread
protection income
Reference portfolio
$1bn notional
receives the Libor spread until maturity and mezzanine investor must bear the subse-
nothing else changes. Using the synthetic quent losses with the corresponding reduc-
CDO described earlier and shown in Figure tion in the mezzanine notional. If the losses
8, consider what happens if one of the ref- exceed 150m, then it is the senior investor
erence entities in the reference portfolio who takes the principal losses.
undergoes the first credit event with a 30% The mechanics of a standard synthetic
recovery, causing a 7m loss. CDO are therefore very simple, especially
The equity investor takes the first loss of compared with traditional cash flow CDO
7m, which is immediately paid to the orig- waterfalls. This also makes them more easi-
inator. The tranche notional falls from 50m ly modelled and priced.
to 43m and the equity coupon, set at
1500bp, is now paid on this smaller notion- The CDO tranche spread
al. These coupon payments therefore fall The synthetic CDO spread depends on a
from 7.5m to 15% times 43m = 6.45m. number of factors. We list the main ones and
If traded in a funded format, the 3m describe their effects on the tranche spread.
recovered on the defaulted asset is either
reinvested in the portfolio or used to reduce Attachment point: This is the amount of
the exposure of the senior-most tranche subordination below the tranche. The
(similar to early amortisation of senior higher the attachment point, the more
tranches in cash flow CDOs). defaults are required to cause tranche
The senior tranche notional is decreased by principal losses and the lower the tranche
3m to 847m, so that the sum of protect- spread.
ed notional equals the sum of the collateral
notionals which is now 990m. This has no Tranche width: The wider the tranche
effect on the other tranches. for a fixed attachment point, the more
This process repeats following each cred- losses to which the tranche is exposed.
it event. If the losses exceed 50m then the However, the incremental risk ascending
the capital structure is usually declining the portfolio loss distribution. We can expect
and so the spread falls. to observe one of the two shapes shown in
Figure 10. They are (i) a skewed bell curve; (ii)
Portfolio credit quality: The lower the a monotonically decreasing curve.
quality of the asset portfolio, measured The skewed bell curve applies to the case
by spread or rating, the greater the risk of when the correlation is at or close to zero. In
all tranches due to the higher default this limit the distribution is binomial and the
probability and the higher the spread. peak is at a loss only slightly less than the
expected loss.
Portfolio recovery rates: The expected As correlation increases, the peak of the
recovery rate assumptions have only a distribution falls and the high quantiles
secondary effect on tranche pricing. This increase: the curves become monotonically
is because higher recovery rates imply decreasing. We see that the probability of
higher default probabilities if we keep the larger losses increases and, at the same
spread fixed. These effects offset each time, the probability of smaller losses also
other to first order. increases, thereby preserving the expected
loss which is correlation independent (for
Swap maturity: This depends on the further discussion see Mashal, Naldi and
shapes of the credit curves. For upward Pedersen (2003)).
sloping credit curves, the tranche curve For very high levels of asset correlations
will generally be upward sloping and so (hardly ever observed in practice), the distri-
the longer the maturity, the higher the bution becomes U-shaped. At maximum
tranche spread. default correlation all the probability mass is
located at the two ends of the distribution.
Default correlation: If default correlation The portfolio either all survives or it all
is high, assets tend to default together defaults. It resembles the loss distribution
and this makes senior tranches more of a single asset.
risky. Assets also tend to survive togeth-
er making the equity safer. To understand Figure 10. Portfolio loss distribution
this more fully we need to better under- for a large portfolio at 0%, 20% and
stand the portfolio loss distribution. 95% correlation
40
The portfolio loss distribution 35
No matter what approach we use to gener- 30 =0
Probability (%)
10
15
47
20
benefit from increasing correlation.
15 The mezzanine tranche becomes more
risky at 50% correlation. As we see in Figure
10
12, the 100% loss probability jumps from
5 0.50% to 3.5%. In most cases mezzanine
investors benefit from falling correlation
0
they are short correlation. However, the cor-
0
100
10
20
30
40
50
60
70
80
90
fall in value.
Mezzanine tranche loss (%)
risk parameters and so have adopted model Figure 13. Senior tranche loss
based approaches. These are discussed on distribution for correlations of 20% and
page 43. 50%. We have eliminated the zero loss
peak, which is greater than 96% in
Customised synthetic CDO tranches both cases
Customisation of synthetic tranches has 0.8
become possible with the fusion of deriva- 0.7
tives technology and credit derivatives. 0.6
Probability (%)
= 20% = 50%
Unlike full capital structure synthetics, which 0.5
issue the equity, mezzanine and senior parts 0.4
of the capital structure, customised synthet- 0.3
ics may issue only one tranche. There are a 0.2
number of other names for customised CDO 0.1
tranches, including bespoke tranches, and 0.0
10
20
30
40
0
single tranche CDOs.
Senior tranche loss (%)
The advantage of customised tranches is
that they can be designed to match exactly
the risk appetite and credit expertise of the Figure 14. Correlation dependence
investor. The investor can choose the credits of CDO tranches
in the collateral, the trade maturity, the
attachment point, the tranche width, the rat- 40
ing, the rating agency and the format (fund- 30
Equity
Mezzanine
m)
of CDS on Name 1
of CDS on Name 2
Investor
of CDS on Name 3 Spread
Lehman
Brothers Bespoke tranche
Contingent
payment
of CDS on Name 99 Reference pool
100 investment grade
of CDS on Name 100 names in CDS format
$10m x 100 assets = $1bn
imagine a queue of all of the credits sorted to default early and hit the equity tranche,
in the order in which they should default. and also because the CDS will have a high-
This ordering will depend mostly on the er spread sensitivity and so require a small-
spread of the asset relative to the other er notional.
credits in the portfolio and its correlation rel- To show this we take an example CDO
ative to the other assets in the portfolio. If with 100 credits, each $10m notional. It has
the asset whose delta you are calculating is three tranches: a 5% equity, a 10% mezza-
at the front of this queue, it will be most like- nine and an 85% senior tranche. The asset
ly to cause losses to the equity tranche and spreads are all 150bp and the correlation
so will have a high delta for the equity between all the assets is the same.
tranche. If it is at the back of the queue then The sensitivity of the delta to changing the
its equity delta will be low. As it is most like- spread of the asset whose delta we are cal-
ly to default after all the other asset, it will be culating is shown in Figure 16. If the single
most likely to hit the senior tranche. As a asset spread is less than the portfolio aver-
result the senior tranche delta will rise. This age of 150bp, then it is the least risky asset.
framework helps us understand the direc- As a result, it would be expected to be the
tionality of delta. last to default and so most likely to impact
The actual magnitude of delta is more dif- the senior-most tranche. As the spread of
ficult to quantify because it depends on the the asset increases above 150bp, it
tranche notional and the contractual becomes more likely to default before the
tranche spread, as well as the features of others and so impacts the equity or mezza-
the asset whose delta we are examining. nine tranche. The senior delta drops and the
For example the delta for a senior tranche equity delta increases.
to a credit whose CDS spread has widened In Figure 17 we plot the delta of the asset
will fall due to the fact that it is more likely versus its correlation with all of the other
Tranche delta
nine tranches are more exposed. For low 5 Mezzanine
4 Senior
correlations, if it defaults it will tend to do so
by itself while the rest of the portfolio tends 3
2
to default together. As a result, the equity
1
tranche is most exposed.
0
There is also a time effect. Through time,
1,000
100
200
300
400
500
600
700
800
900
0
senior and mezzanine tranches become
Asset spread (bp)
safer relative to equity tranches since less
time remains during which the subordina-
tion can be reduced resulting in principal Idiosyncratic versus systemic risk
losses. This causes the equity tranche delta In terms of how they are exposed to credit,
to rise through time while the mezzanine there is a fundamental difference between
and senior tranche deltas fall to zero. equity and senior tranches. Equity tranches
Building intuition about the delta is not triv- are more exposed to idiosyncratic risk they
ial. There are many further dependencies to incur a loss as soon as one asset defaults.
be explored and we intend to describe these The portfolio effect of the CDO is only
in a forthcoming paper. expressed through the fact that it may take
several defaults to completely reduce the
Higher order risks equity notional. This implies that equity
If properly hedged, the dealer should be investors should focus less on the overall
insensitive to small spread movements. properties of the collateral, and more on try-
However, this is not a completely risk-free ing to choose assets which they believe will
position for the dealer since there are a num-
ber of other risk dimensions that have not Figure 17. Dependency of tranche
been immunised. These include correlation delta on the assets correlation with the
sensitivity, recovery rate sensitivity, time rest of the portfolio
decay and spread gamma. There is also a
10.0
risk to a sudden default which we call the 9.0
value-on-default risk (VOD). 8.0
Tranche delta
7.0
For this reason, dealers are motivated to 6.0
5.0
do trades that reduce these higher order 4.0
risks. The goal is to flatten the risk of the 3.0
2.0
correlation book with respect to these high- 1.0
0.0
er order risks either by doing the offsetting 0 10 20 30 40 50
trade or by placing different parts of the Correlation with rest of portfolio (%)
capital structure with other buyers of cus- Equity Mezzanine Senior
tomised tranches.
not default. As a result we would expect will usually receive a fixed coupon through-
equity tranche buyers to be skilled credit out the life of the transaction and any upside
investors, able to pick the right credits for or remainder in the reserve account at matu-
the portfolio, or at least be able to hedge the rity. If structured to build to a predetermined
credits they do not like. level, the equity tranche will usually receive
On the other hand, the senior investor has a excess interest only after the reserve
significant cushion of subordination to insu- account is fully funded. More details are
late them from principal losses until maybe provided in Ganapati and Ha (2002).
20 or more of the assets in the collateral have Other structures incorporated features to
defaulted. As a consequence, the senior share some of the excess spread with
investor is truly taking a portfolio view and so mezzanine holders or to provide a step
should be more concerned about the average up coupon to mezzanines if losses exceed-
properties of the collateral than the quality of ed a certain level or if the tranche was
any specific asset. The senior tranche is real- downgraded. Finally, over-collateralisation
ly a deleveraged macro credit trade. trigger concepts were adopted from cash
flow CDOs.
Evolution of structures
Initially full capital structure synthetic CDOs Principal protected structures
had almost none of the structural features Investors who prefer to hold highly rated
typically found in other securitised asset assets can do so by purchasing CDO tranch-
classes and cash flow CDOs. It was only in es within a principal protected structure.
1999 that features that diverted cash flows This is designed to guarantee to return the
from equity to debt holders in case of cer- investors initial investment of par. One par-
tain covenant failures began entering the ticular variation on this theme is the Lehman
landscape. The intention was to provide Brothers High Interest Principal Protection
some defensive mechanism for mezzanine with Extendible Redemption (HIPER). This is
holders fearing that the credit cycle would typically a 10-year note which pays a fixed
affect tranche performance. Broadly, these coupon to the investor linked to the risk of a
features fit into two categories ones that CDO equity tranche.
build extra subordination using excess This risk is embedded within the coupons
spread, and others that use excess spread of the note such that each default causes a
to provide upside participation to mezza- reduction in the coupon size. However the
nine debt holders. investor is only exposed to this credit risk for
The most common example of structural a first period, typically five years, and the
ways to build additional subordination is the coupon paid for the remaining period is
reserve account funding feature. Excess frozen at the end of year five. The coupon is
spread (the difference between premium typically of the form:
received from the CDS portfolio and the
tranche liabilities) is paid into a reserve Portfolio loss
account. This may continue throughout the Coupon = 8% max 1 , 0
Tranche size
life of the deal or until the balance reaches a
predetermined amount. If structured to
accumulate to maturity, the equity tranche In Figure 18 we show the cash flows
income to offset some of the cost of pro- Hedge funds have been the main growth
tection on the widest names. user of credit options, using them for credit
2. The investor may buy CDO equity and arbitrage and also for debt-equity strategies.
delta hedge. The net positive gamma They are typically buyers of volatility, hedg-
makes this trade perform well in high ing in the CDS market and exploiting the
spread volatility scenarios. By dynamical- positive convexity. Asset managers seeking
ly re-hedging, the investor can lock in this to maximise risk-adjusted returns are
convexity. The low liquidity of CDOs involved in yield-enhancing strategies such
means that this hedging must continue as covered call writing. Bank loan portfolio
to maturity. managers are beginning to explore default
3. The investor may use the carry from CDO swaptions as a cheaper alternative to buying
equity to over-hedge the whole portfolio, outright credit protection via CDS.
creating a cheap macro short position. One source of credit optionality is the cash
While this is a negative carry trade, it can market. Measured by market value weight,
be very profitable if the market widens 5.6% of Lehman Brothers US Credit Index
dramatically or if a large number of and 54.7% of Lehman Brothers US High
defaults occur. Yield Index have embedded call or put
options. Hence, two strategies which have
For more details see Isla (2003). been, and continue to be important in the
bond options market are the repack trade
Credit options and put bond stripping.
Activity in credit options has grown sub-
stantially in 2003. From a sporadic market The repack trade
driven mostly by one-off repackaging deals, The first active market in credit bond
it has extended to an increasingly vibrant options was developed in the form of the
market in both bond and spread options, repack trade, spearheaded by Lehman
options on CDS and more recently options Brothers and several other dealers. Figure
on portfolios and even on CDO tranches. 20 (overleaf) shows the schematic of one
This growth of the credit options market such transaction.
has been boosted by declines in both In a typical repack trade, Lehman Brothers
spread levels and spread volatility. The purchased $32,875,000 of the Motorola
reduction in perceived default risk has (MOT) 6.5% 2028 debentures and placed
made hedge funds, asset managers, insur- them into a Lehman Trust called CBTC. The
ers and proprietary dealer trading desks Trust then issued $25 par class A-1
more comfortable with the spread volatility Certificates to retail investors with a coupon
risks of trading options and more willing to set at 8.375% the prevailing rate for MOT in
exploit their advantages in terms of lever- the retail market at the time. Since the
age and asymmetric payoff. 8.375% coupon on the CBTC trust is higher
The more recent growth in the market for than the coupon on the MOT Bond, the CBTC
options on CDS has also been driven by the trust must be over-collateralised with enough
increased liquidity of the CDS market, face value of MOT bonds to pay the 8.375%
enabling investors to go long or short the coupon. An A-2 Principal Only (PO) tranche
option delta amount. captures the excess principal. Both class A-1
and class A-2 certificates are issued with an This option can be viewed as an extension
embedded call. option since by failing to exercise it, the bond
This call option was sold separately to maturity is extended. In the past several
investors in a form of a long-term warrant. years, the market has priced these bonds as
The holder of the MOT call warrant has the though they matured on the first put date and
right but not the obligation to purchase the has not given much value to the extension
MOT bonds from the CBTC trust beginning option. Recently, credit investors have
on 19/7/07 and thereafter at the preset call realised a way to extract this extension risk
strike schedule. This strike is determined by premium via a put bond stripping strategy.
the proceeds needed to pay off the A-1 cer- Essentially, an investor can buy the put
tificate at par plus the A-2 certificate at the bond, and sell the call option to the first put
accreted value of the PO. Because retail date at a strike price of par. Thus, the investor
investors are willing to pay a premium for has a long position in the bond coupled with a
the par-valued low-notional bonds of well- synthetic short forward (long put plus short
known high quality issuers, the buyers of the call) with a maturity coinciding with the first
call warrant can use this structure to source put date. He then hedges this position by
volatility at attractive levels. asset swapping the bond to the put date,
effectively eliminating all of the interest rate
Put bond stripping risk and locking in the cheap volatility. Given
According to Lehman Brothers US Credit the small amount of outstanding put bonds,
Index, bonds with embedded puts consti- this strategy has led to more efficient pricing
tute approximately 2.3% of the US credit of the optionality in these securities.
bond market, by market value. These bonds
grant the holder the right, but not the obli- Bond options
gation to sell the bond back to the issuer at There are a variety of bond options traded in
a predetermined price (usually par) at one or the market. The two most important ones
more future dates. for investors are:
ket is that protection calls (option to buy pro- for the right, but not the obligation, to sell
tection) are called payer default swaptions. CDS protection on a reference entity at a
Protection puts (option to sell protection) are predetermined spread on a future date. This
called receiver default swaptions. spread is the option strike.
Unlike price options on bonds, the exercise We do not need to consider what happens
decision for default swaptions is based on if the reference entity experiences a credit
credit spread alone. As a result, they are event between trade date and expiry date as
essentially a pure credit product, with pricing they would never exercise the option in this
being mostly driven by CDS spread volatility. case. As a result, there is no need for a
Default swaptions give investors the oppor- knockout feature for receiver default swap-
tunity to express views on the future level and tions. Consider the following example.
variability of default swap spreads for a given Lehman Brothers pays 1.20% for an at-
issuer. They can be traded outright or embed- the-money receiver default swaption on
ded in callable CDS. The typical maturity of five-year GMAC, struck at the current five
the underlying CDS is five years but can range year spread of 265bp and with three
from one10 years, and the time to option months to expiry. The investor is short the
expiry is typically three months to one year. option. From the investors perspective, the
relevant scenarios are:
Payer default swaption
The option buyer pays a premium to the If five-year GMAC trades above 265bp in
option seller for the right but not the obliga- three months, Lehman does not exercise,
tion to buy CDS protection on a reference as they can sell protection for a higher
entity at a predetermined spread on a future spread in the market. The investor has
date. Payer default swaptions can be struc- realised an option premium of 1.20% in a
tured with or without a provision for knock quarter of a year.
out at no cost if there is a credit event
between trade date and expiry date. If the If five-year GMAC trades at 238bp in three
knock out provision is included in the swap- months, the trade breaks even. (1.20% up
tion, the option buyer who wishes to main- front option premium equals the payoff of
tain protection over the entire maturity range (265bp238bp)=27bp times the five-year
can separately buy protection on the under- PV01 of 4.39). If five-year GMAC trades
lying name until expiry of the swaption. below 238bp in three months, the loss on
The relevant scenarios for this investment the exercise of the option will be greater
are complementary to the ones in the case than the upfront premium and the investor
of the protection put. If spreads tighten by will underperform on this trade.
the expiry date, the option buyer will not
exercise the right to buy protection at the Hedging default swaptions
strike and the option seller will keep the Dealers hedge these default swaptions using
option premium. a model of the type discussed on page 49.
The underlying in a default swaption is the for-
Receiver default swaption ward CDS spread from the option expiry date
In a receiver default swaption, the option to the maturity date of the CDS. Theoretically,
buyer pays a premium to the option seller a knock-out payer swaption should be delta
payer default swaption will exercise the Clean and perfect asset swaps
option and lock in the portfolio protection One important theme is the isolation of the
at more favourable levels. Conversely, if the pure credit risk component in a given
TRAC-X spread is tighter than the strike, the instrument. For example, a European CDO
holder of the receiver swaption will benefit investor may wish to access USD collateral
from exercising the option and realising the without incurring any of the associated cur-
MTM gain. rency risks.
Investors can monetise a view on the future Cross-currency asset swaps are the tradi-
range of market spreads by trading bearish tional mechanism by which credit investors
spread (buying at-the-money receiver swap- transform foreign currency fixed-rate bonds
tion and selling farther out-of-money receiver into local currency Libor floaters. This has
swaption) or bullish spread (buying ATM payer the benefit that it substantially reduces the
swaption and selling farther out-of-money currency and interest rate risk, converting
payer swaption) strategies. Other strategies the bond from an FX, interest rate and cred-
include expressing views on spread changes it play into an almost pure credit play.
over a given time horizon by trading calendar However, the currency risk has not been
spreads (buying near maturity options and completely removed. First, note that a cross
selling farther maturity options). currency asset swap is really two trades: (i)
Finally, because the TRAC-X spread is less purchase of a foreign currency asset; and (ii)
subject to idiosyncratic spread spikes, and entry into a cross-currency swap. In the case
because of the existing two-way markets of a European investor purchasing a dollar
with varying strikes, investors can express asset, the investor receives Euribor plus a
their views on the direction of changes in spread paid in euros.
the macro level of spread volatility by trading As long as the underlying dollar asset
straddles, ie, simultaneously buying payer does not default during the life of the asset
and receiver default swaptions as a way to swap there is no currency risk to the
go long volatility while being neutral to the investor. However, if the asset does
direction of spread changes. default, the investor loses the future dollar
coupons and principal of the asset, just
Hybrid products receiving some recovery amount which is
Hybrid credit derivatives are those which paid in dollars on the dollar face value. As
combine credit risk with other market risks the cross-currency swap is not contingent,
such as interest rate or currency risk. meaning that the payments on the swap
Typically, these are credit event contingent contract are unaffected by any default of
instruments linked to the value of a deriva- the asset, the investor is therefore obliged
tives payout, such as an interest rate swap to either continue the swap or to unwind it
or an FX option. at the market value with a swap counter-
There are various motivations for entering party. This unwind value can be positive or
into trades which have these hybrid risks. negative the investor can make a gain or
Below, we give an overview of the economic loss depending on the direction of move-
rationale for different types of structures. We ments in FX and interest rates since the
discuss the modelling of hybrid credit deriva- trade was initiated.
tives in more detail on page 51. The risk is significant. We have modelled
Probability (%)
the defaulted credit.
4
As a result, the basic cross-currency
asset swap has a default contingent inter- 3
To determine the spread we therefore need for the risk of default per small time interval.
to be able to value the protection and pre- Within this model the interest rate depen-
mium legs. It is important to take into dency drops out.
account the timing of the credit event Given a CDS which has a flat spread curve
because this can have a significant effect on at 150bp, and assuming a 50% recovery
the present value of the protection leg and rate, the implied hazard rate is 0.015 divided
also the amount of premium paid on the pre- by 0.5, which implies a 3% hazard rate. The
mium leg. Within the hazard rate approach implied one-year survival probability is there-
we can solve this timing problem by condi- fore exp(0.03)=97.04%. For two years it is
tioning on the probability of defaulting with- exp(0.06)=94.18%, and so on.
in each small time interval [s, s+ds], given by
Q(0,s)(s)ds, then paying (1R) and discount- Valuation of a CDS position
ing this back to today at the risk-free rate. The value of a CDS position at time t follow-
Assuming that the hazard rate and risk free ing initiation at time 0 is the difference
rate term structures are flat, we can write between the market implied value of the
the value for the protection leg as protection and the cost of the premium pay-
ments, which have been set contractually at
SC. We therefore write
(1 R)(1 e ( r + )T )
T
(1 R) e ( r + ) s ds =
0
r+ MTM(t) = (Protection PV Premium PV),
The value of the premium leg is the PV of the where the sign is positive for a long protec-
spread payments which are made to default tion position and negative for a short protec-
or maturity. If we assume that the spread S tion position. If the current market spread is
on the premium leg is paid continuously, we given by S(t) then the MTM can be written as
can write the present value of the premium
leg as MTM (t ) = ( S (t ) S (0)) RPV01
( r + )T
S (1 e
T
)
S e ( r + ) s ds = where the RPV01 is the risky PV01 which is
0
r + given by
(1 e ( ),
Equating the protection and premium legs
r + )(T t )
and solving for the breakeven spread gives
RPV 01(t ) =
(r + )
S = (1 R ) and where
S (t )
This relationship is known as the credit tri- = .
angle because it is a relationship between 1 R
three variables where knowledge of any two
is sufficient to calculate the third. It basical- An investor buys $10m of five-year protec-
ly states that the spread paid per small time tion at 100bp. One year later, the credit
interval exactly compensates the investor trades at 250bp. Assuming a recovery rate
of 40%, the value is given by substituting, expected recovery rate following a credit
r=3.0%, R=40%, S(t)=0.025, S(0)=0.01 and event where the expectation is under the
t=4 into the above equation to give risk-neutral measure.
=4.17% and an MTM value of $521,661. Such expectations are only available from
This is a simple yet fairly accurate model price information, and the problem in credit
which works quite well when the interest is that given one price, it is difficult to sepa-
rate and credit curves are flat. When this is rate the probability of default from the recov-
not the case, it becomes necessary to use ery rate expectation.
bootstrapping techniques to build a full term The market standard is therefore to revert to
structure of hazard rates. This may be rating agency default studies for estimates of
assumed to be piecewise flat or piecewise recovery rates. These typically show the aver-
linear. For a description of such a model see age recovery rate by seniority and type of
OKane and Turnbull (2003). credit instrument, and usually focus on a US
corporate bond universe. Adjustments may
Default probabilities be made for non-US corporate credits and for
The default probabilities calculated for certain industrial sectors.
pricing purposes can be quite different Problems with rating agency recovery
from those calculated from historical statistics include the fact that they are back-
default rates of assets with the same rat- ward looking and that they only include the
ing. These real-world default probabilities default and bankruptcy credit events
are generally much lower. The reason for restructuring is not included. In their favour,
this is that the credit spread of an asset one should say that, as they represent the
contains not just a compensation for pure price of the defaulted asset as a fraction of
default risk; it also depends on the mar- par some 30 days after the default event,
kets risk aversion expressed through a risk they are similar to the definition of the recov-
premium, as well as on supply-and- ery value in a CDS.
demand imbalances. Recent work (Altman et al 2001) shows that
One should also comment on the markets there is a significant negative correlation
use of Libor as a risk-free rate in pricing. between default and recovery rates. One
Pricing theory shows that the price of a way to incorporate this effect is to assume
derivative is the cost of replicating it in a risk- that recovery rates are stochastic. The stan-
free portfolio using other securities. Since dard approach is to use a beta distribution.
most market dealers are banks which fund
close to Libor, the cost of funding these Modelling default correlation
other securities is also close to Libor. As a By modelling correlation products, we mean
consequence it is the effective risk-free rate modelling products whose pricing depends
for the derivatives market. upon the joint behaviour of a set of credit
assets. These include default baskets and
Calibrating recovery rates synthetic CDOs. As a result of the growth in
The calibration of recovery rates presents a usage of these products, this is an area of
number of complications for credit deriva- pricing which has recently gained a lot of
tives. Strictly speaking, the recovery rate attention, and in which we have seen a num-
used in the pricing of credit derivatives is the ber of significant modelling developments.
4
simulated 1,000 pairs of asset returns mod-
3
2
3
elled as normally distributed random vari-
2
1
VA 1
VA ables, for two firms i and j for two different
0
0
-4 -2
-1
0 2 4
-4 -2
-1
0 2 4 asset return correlations of 10% and 90%.
CB CB
-2
-3
-2
-3
The vertical and horizontal lines represent
-4
Both firms default
-4 default thresholds for firms i and j respec-
tively. Clearly we see that the probability of
both i and j defaulting, represented by the
number of points in the bottom left quad-
In this section we will describe some of rant defined by the default thresholds,
these current models, show how they can increases as the asset return correlation
be applied to the valuation of baskets and increases. Therefore asset correlation
CDOs, and towards the end discuss model leads to default correlation.
calibration issues. Although the pay-offs of multi-credit
default-contingent instruments such as nth-
Modelling joint defaults to-default baskets and synthetic loss tranch-
The valuation of default-contingent instru- es cannot be statically replicated by trading
ments calls for the modelling of default in a set of single-credit contracts, the cur-
mechanisms. As discussed earlier, a classical rent market practice is to value correlation
dichotomy in credit models distinguishes products using standard no-arbitrage argu-
between a structural approach, where ments. It follows that the valuation of these
default is triggered by the market value of the multi-credit exposures boils down to the
borrowers assets (in terms of debt plus equi- computation of (risk-neutral) expectations
ty) falling below its liabilities, and a reduced- over all possible default scenarios.
form approach, where the default event is A number of different hybrid frameworks
directly modelled as an unexpected arrival. have been proposed in the literature for mod-
Although both the structural and the reduced- elling correlated defaults and pricing multi-
form approaches can in principle be extended name credit derivatives. Hull and White (2001)
to the multivariate case, structural models generate dependent default times by diffus-
calibrated to market-implied default probabili- ing correlated asset values and calibrating
ties (often called hybrid models) have gained default thresholds to replicate a set of given
favour among practitioners because of their marginal default probabilities. Multi-period
tractability in high dimensions. extensions of the one-period CreditMetrics
If we think of defaults as generated by paradigm are also commonly used, even if
asset values falling below a given boundary, they produce the undesirable serial indepen-
then the probabilities of joint defaults over a dence of the realised default rate.1 While most
specified horizon must follow from the joint multi-credit models require simulation, the
need for accurate and fast computation of such as the simulation approach and the
greeks has pushed researchers to look for semi-analytical framework described below,
modelling alternatives. Finger (1999) and, use the dependence among asset returns to
more recently, Gregory and Laurent (2003) generate joint defaults, therefore avoiding
show how to exploit a low-dimensional factor the need for a direct estimation of joint
structure and conditional independence to default probabilities.
obtain semi-analytical solutions.
Default-time simulation
Asset and default event correlation Monte Carlo models generally aim at the
Default event correlation (DEC) measures generation of default paths, where each
the tendency of two credits to default joint- path is simply a list of default times for each
ly within a specified horizon. Formally, it is of the credits in the reference portfolio
defined as the correlation between two drawn at random from the joint default dis-
binary random variables that indicate tribution. Knowing the time and identity of
defaults, ie each default event allows for a precise valu-
ation of any multi-credit product, no matter
p AB p A p B how complex the contractual specification
DEC = of the payoff.
p A (1 p A ) p B (1 p B ) In an influential paper, Li (2000) presents a
simple and computationally inexpensive
where pA and pB are the marginal default algorithm for simulating correlated defaults.
probabilities for credits A and B, and PAB is His methodology builds on the implicit
the joint default probability. Of course, pA, pB assumption that the multivariate distribution
and pAB all refer to a specific horizon. Notice of default times and the multivariate distri-
that default event correlation increases lin- bution of asset returns share the same
early with the joint probability of default and dependence structure, which he assumes to
is equal to zero if and only if the two default be Gaussian and is therefore fully charac-
events are independent. Its limits are not terised by a correlation matrix.
100% to +100% but are actually a function For valuation purposes, we need to sample
of the marginal probabilities themselves. from the multivariate distribution of default
Default event correlations are the funda- times under the risk-neutral probability mea-
mental drivers in the valuation of multi-name sure. In this case, it is common practice to
credit derivatives. Unfortunately, the scarcity back out the marginal distributions of default
of default data makes joint default probabili- times, which we will denote with F1,F2,,Fd,
ties, and thus default event correlations, from single-credit defaultable instruments
very hard to estimate directly. As a result, (such as CDS). We then join these marginal
market participants rely on alternative meth- distributions with a correlation matrix, which
ods to calibrate the frequency of joint according to the stated assumptions repre-
defaults within their models. Hybrid models, sents the correlation matrix of the asset
returns of the reference credits. Since asset
1
Finger (2000) offers an excellent comparison of several returns are not directly observable, it is com-
multivariate models in terms of the default distributions that
they generate over time when calibrated to the same
mon practice to proxy asset correlations
marginals and first-period joint default probabilities. using equity correlations. Towards the end of
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
6 4 2 0 2 4 6 0 2 4 6 8 10
x Time (years)
this section we will discuss whether this 1. Choleski decomposition of the correla-
seems to be a reasonable approximation. tion matrix to simulate a multivariate
The high-dimensionality of multi-credit Normal random vector X R d with cor-
instruments means that it is not possible to relation .
use market prices to obtain the full depen- 2. Transform the vector into the unit hyper-
dence structure. Instead, practitioners gen- cube using
erally estimate the necessary correlations
using historical returns, implicitly relying on
U = (N (x1 ), N (x 2 ) ,..., N (x d ))
the extra assumption that the correlations
among asset returns remain unchanged
when we move from the objective to the 3. Translate U into the corresponding
pricing probability measure. default times vector t using the inverse of
In the bivariate case, the joint distribution the marginal distributions:
function of default times and is simply
given by
= ( 1 , 2 ,..., d ) = (F11(u1), F21(u 2 ),..., Fd1 (u d ))
E[f ()] where = (1,2,...,d) represents the arises whether some other numerical
vector of default times and f is a function approach can be used.
describing the discounted cash flows (both
positive and negative) associated with the A semi-analytical approach
instrument under consideration (see, eg, The recent development of correlation trad-
Mashal and Naldi (2002b)). ing and the associated need for fast compu-
In summary, Monte Carlo simulation tation of sensitivities have generated a great
allows for accurate valuations and risk mea- deal of interest in semi-analytical models. To
surements of multi-credit payoffs, even enjoy the advantages of fast pricing, one
when complex path-dependencies such as needs to impose more structure on the
reserve accounts, interest coverage or col- problem. One way to do this is to rely on
lateralisation tests are involved. This is two basic simplifications:
because both the exact default times and
the identity of the defaulters are known on 1. assume a one-factor correlation structure
every simulated path. Moreover, we can for asset returns, and
easily expand the set of variables to be 2. discretise the time-line to allow for a finite
treated as random. For example, Frye (Risk, set of dates at which defaults can hap-
2003) argues that modelling stochastic pen. These are chosen with a resolution
recoveries and allowing for negative to provide a sufficient level of accuracy.
correlation between recovery and default
rates is essential for a proper valuation of In particular, 1) makes it possible to com-
credit derivatives. pute the risk-neutral loss distribution of the
The precision and flexibility of this reference portfolio for any given horizon,
approach, however, come at the cost of while 2) is needed to price an instrument
computational speed. The basic problem of knowing only the loss distribution of the
using simulation is that defaults are rare reference portfolio at a finite set of dates.
events, and a large number of simulation While these two assumptions are suffi-
paths are usually required to achieve a suf- cient to price plain vanilla portfolio swaps,
ficient sampling of the probability space. derivatives structures with more complex
There are ways to improve the situation. path-dependencies may also require that
Useful techniques include antithetic sam- we approximate the payoff function f().
pling, importance sampling and the use of Even in this case, it is usually possible to
low-discrepancy sequences. This problem obtain reasonable approximations, so that
becomes particularly significant when we the benefit of precise sensitivities can be
turn our attention to the calculation of sen- retained at a relatively low cost. Moreover,
sitivity measures, since for a reasonable the error can be controlled by comparing
number of paths the simulation noise can the analytical solution to a Monte Carlo
be similar to or greater than the price implementation.
change due to the perturbation of the input We now discuss how to exploit a one-factor
parameter. Once again, techniques exist to model to construct the risk neutral loss distri-
alleviate the problem, but it is hard to bution. Later on, we will show how to use a
achieve precise hedge ratios in a reason- sequence of loss distributions at different
able amount of time. The question then horizons to price synthetic loss tranches.
lated to multiple periods for each of the can be computationally slow if there are a
assets in the basket. These asset values large number of assets in the portfolio.
may be correlated internally in order to Another way is to use a semi-analytical
induce a default correlation. If the asset approach which relies on the fact that a
value falls below the default threshold at standard synthetic loss tranche can be
the future period, the asset defaults and a priced directly from its loss distribution.
recovery rate is drawn from a beta distribu- To see this, consider a portfolio CDS on a
tion. The recovery amount and asset value tranche defined by the attachment point Kd
can be correlated. The model is calibrated and the upper boundary Ku, expressed as
using historical default statistics and the percentages of the reference portfolio
assigned rating is linked to the expected notional. This is Nport so that the tranche
loss of the basket to the trade maturity. notional is given by
S&P has recently switched its approach
from a weakest-link approach which assigns N tranche = N port ( K u K d )
an FTD the rating of the lowest-rated entity
in the basket, to one based on a Monte Carlo
model, which uses the same framework as The maturity of the portfolio swap is T, and
its CDO Evaluator model. for each time tT we denote by L(t) the
cumulative portfolio loss up to time t. The
Pricing synthetic loss tranches tranche loss is therefore given by
We would now like to consider in detail
how we would set about valuing a loss
tranche. One approach is to use the times Ltranche (t ) = max[L(t ) K d ,0] max[L(t ) K u ,0]
to default model via a Monte Carlo imple-
mentation as discussed earlier. Given that Note that this is similar to an option style
we know when each asset in the portfolio payoff. Indeed it is possible to think of CDO
defaults in each simulation path, we would tranches as options on the portfolio loss
proceed as follows: amount.
The two legs of the swap can be priced in
1. Calculate the present value of all principal the same way as a CDS if we introduce a
losses on the protection leg in each path. tranche default probability P(t) defined as
2. Calculate the PV of 1bp on the premium
leg taking care to reduce the notional of
E 0Q [ Ltranche (t )]
the tranche following defaults which P(t ) =
cause principal losses in that path. N tranche
3. Average both the 1bp premium leg PV,
known as the tranche PV01, and the pro- where we use the risk-neutral (pricing) mea-
tection leg PV over all paths. sure for taking the expectation.
4. Divide the protection PV by the tranche Assuming that the credit, recovery rate and
PV01 to compute the breakeven tranche interest rates processes are independent,
spread. the contingent leg is therefore given by
K An asymptotic approximation
N tranche B(0, t i )( P (t i ) P(t i 1 )) It is possible to exploit the high dimension-
i =1
ality of the CDO to derive a closed form
model for analysing CDO tranches. In fact
To value the premium leg, we denote the we can use this approach to obtain results
contractual spread on the tranche by s, and which differ from the exact approach by
the coupon payment dates as 15% for a real CDO.
0=T0<T1<<Tk=T. The accrual factor for To begin with, let us assume that the port-
the ith payment is denoted by i. The PV of folio is homogeneous, ie, that each assets
the premium leg is then given by and default probability are the same. Hence,
all assets have the same default threshold C.
Premium Leg PV= As a result, the conditional default probabili-
ty of any individual issuer in the reference
n portfolio is given by
sN tranche j (1 P(T j )) B(0, T j )
j =1
C Z
p(Z M ) = N M .
The PV of the tranche from the perspective 1 2
of the investor who is receiving the spread is
therefore given by
If we also assume that the loss exposure to
Tranche PV = Premium Leg PV Protection each issuer is of the same notional amount
Leg PV. u, the probability that the percentage loss L
of the portfolio is ku is equal to the proba-
As a result we can value a standard syn- bility that exactly k of the m issuers default,
which is given by the simple binomial the tranche notional this is given by
k
m C ZM max(L K1 , 0) max (L K 2 , 0)
P[L = ku | Z ]= N L(K1 , K 2 ) =
K 2 K1
k 1 2
mk
1 N C Z M If K<1, then
1 2
L(K1 , K 2 ) K L K1 + K (K 2 K1 )
This distribution becomes computationally This equation shows that we can easily derive
intensive for large values of m, but we can the loss distribution of the tranche from that
use methods of varying sophistication to of the reference portfolio. This is discontinu-
approximate it. One very simple and sur- ous at the edges owing to the probability of
prisingly accurate method is the so-called the portfolio loss falling outside the interval,
large homogeneous portfolio (LHP) and this discontinuity becomes more pro-
approximation, originally due to Vasicek nounced when the tranche is narrowed. We
(1987). Since the asset returns, conditional can further compute the expected loss of the
on ZM are independent and identically dis- tranche analytically. For more details see
tributed, by the law of large numbers, the OKane and Schloegl (2001). We then arrive at
fraction of issuers defaulting will tend to
the conditional probability of an asset E [L (K 1 , K 2 )] =
defaulting p(ZM). As a result, the condition- ( ) (
N 2 N 1 (K 1 ),C , 1 2 N 2 N 1 (K 2 ),C , 1 2 )
al loss is directly linked to the value of the K 2 K1
market factor ZM which itself is normally
distributed. We can then write the probabil- This is only a one-period approach.
ity of the portfolio loss being less than or However, it can easily be extended to multi-
equal to some loss threshold K as ple periods as described earlier.
Correlation skew
[
P[L K]= N p 1 ( K ) ] It is now possible to observe tradable
tranche spreads for different levels of
where N(x) is the cumulative normal func- seniority. If we attempt to imply out the mar-
tion. More involved calculations show that ket correlation using a simple Gaussian cop-
the distribution of L actually converges to ula model fitted to observed market tranche
this limit as m tends to infinity. spreads, we can observe a skew in the aver-
We can use the portfolio distribution to age correlation as a function of the width
derive the loss distributions of individual and attachment point of the tranche. This
tranches. If L(K1,K2) is the percentage loss skew may be the consequence of a number
of the mezzanine tranche with attachment of factors such as the assumption of inde-
point K1 and upper loss threshold K2, then pendence of default and recovery rates. It
this can be written as a function of the loss may also be due to our incorrect specifica-
on the reference portfolio. As a fraction of tion of the dependence structure, as
Test Statistic
assets is to implement a univariate struc- pvalue = 0.0001
pvalue = 0.01
credit tool that first computes a measure of
distance-to-default and then maps it into a
default probability (EDF) by means of a 10
0
1 2 3 4 5 6
10 10 10 10 10 10
historical analysis of default frequencies.4 In Null DoF
of asset returns. 10
3
Methodology
A key observation in modelling and testing
Test Statistic
2
10
dependencies is that any d-dimensional pvalue = 0.0001
multivariate distribution can be specified
via a set of d marginal distributions that are 1
10
knitted together using a copula function. pvalue = 0.01
Null DoF
10
4
10
5
10
6
dom (DoF). The latter plays a crucial role in database. The reader should keep in mind,
modelling and explaining extreme co- however, that equity values are observable,
movements in the underlyings. while asset values have been backed out
Moreover, it is well known that the by means of KMVs implementation of a uni-
Student-t distribution is very close to the variate Merton model. We apply our analysis
Gaussian when the DoF is sufficiently large to a portfolio of 30 credits included in the
(say, greater than 30); thus, the Gaussian Dow Jones Industrial Average and use daily
model is nested within the t-family. The data covering the period from 31/12/00 to
same statement holds for the underlying 8/11/02. The reader is referred to Mashal,
dependence structures, and the DoF param- Naldi and Zeevi (2003) for more examples
eter effectively serves to distinguish the two using high yield credits and different sam-
models. This suggests how empirical stud- pling frequencies.
ies might test whether the ubiquitous Following the methodology mentioned
Gaussian hypothesis is valid or not. In par- above, we estimate the number of degrees-
ticular, these studies would target the of-freedom (DoF) of a t-copula without
dependence structure rather than the distri- imposing any structure on the marginal dis-
butions themselves, thus eliminating the tributions of returns. Then, using a likelihood
effect of marginal returns that would con- ratio test statistic, we perform a sensitivity
taminate the estimation problem in the lat- analysis for various null hypotheses of the
ter case. To summarise, the t-dependence underlying tail dependence, as captured by
structure constitutes an important and quite the DoF parameter. The two horizontal lines
plausible generalisation of the Gaussian in Figure 26 represent significance levels of
modelling paradigm, which is our main moti- 99% and 99.99%; a value of the test statis-
vation for focusing on it. tic falling below these lines corresponds to a
With this in mind, the key question that we value of DoF that is not rejected at the
now face is how to estimate the parameters respective significance levels.
of the dependence structure. Mashal, Naldi The minimal value of the test statistic is
and Zeevi (2002) describe a methodology achieved at 12 DoF for asset returns and at
which can be used to estimate the parame- 13 DoF for equity returns. In both cases, we
ters of a t-copula without imposing any can reject any value of the DoF parameter
parametric restriction on the marginal distri- outside the range [10,16] with 99% confi-
butions of returns. They also construct a dence; in particular, the null assumption of
likelihood ratio statistic to test the hypothe- a Gaussian copula (DoF=) can be rejected
sis of Gaussian dependence, and compare with an infinitesimal probability of error.
the dependence structures of asset and Also, the point estimates of the asset
equity returns to evaluate the common prac- returns DoF lies within the non-rejected
tice of proxying the former with the latter. In interval for the equity returns DoF, and vice
the remainder of this section we summarise versa, indicating that the two are essential-
their empirical findings. ly indistinguishable from a statistical signif-
icance viewpoint. Moreover, the difference
Empirical results between the joint tail behaviour of a 12- and
For the purpose of this study, asset and a 13-DoF t-copula is negligible in terms of
equity values are both obtained from KMVs any practical application.
For this model, we have been able to cal- 15-20 $41,645 (1.70) $90,231 (1.62) 117
culate a closed-form solution for the densi-
ty of the portfolio loss distribution, and
show in OKane and Schloegl (2003) that it popular conjecture that the different lever-
possesses the same tail dependent proper- age of assets and equity will necessarily
ties as described above, ie, a widening of create significant differences in their joint
senior spreads and a reduction in equity dynamics seems to be empirically unfound-
tranche spreads. ed. From a practical point of view, these
results represent good news for practition-
Summary ers who only have access to equity data for
Our empirical investigation of the depen- the estimation of the dependence parame-
dence structure of asset returns sheds ters of their models.
some light on the two main issues that
were raised at the beginning of this section. Modelling credit options
First, the assumption of Gaussian depen- We separate credit options into options on
dence between asset returns can be reject- bonds and options on default swaps.
ed with extremely high confidence in favour
of an alternative fat-tailed dependence. Pricing options on bonds
Multivariate structural models that rely on Options on corporate bonds are naturally
the normality of asset returns will generally divided into three groups according to how
underestimate default correlations, and the exercise price is specified. The option
thus undervalue junior tranches and over- can be struck on price, yield or credit
value senior tranches of multi-name credit spread. The exercise price is constant for
products. A fat-tailed dependence of asset options struck on price, but for options
returns will produce more accurate joint struck on yield it depends on the time to
default scenarios and more accurate valua- maturity of the underlying bond and is found
tions. Second, the dependence structures from a standard yield-to-maturity calculation.
of asset and equity returns appear to be Obviously, for European options there is no
strikingly similar. The KMV algorithm that difference between specifying a strike price
produces the asset values used in our anal- and a strike yield.
ysis is nothing other than a sophisticated Bond options struck on spread are differ-
way of de-leveraging the equity to get to the ent. For credit spread options the exercise
value of a companys assets. Therefore, the price depends both on the time to maturity
how to price options to buy or sell protection = 0 are simply ignored in that case and the
through CDS. Bond options struck on credit distribution of ST will be such that the proba-
spread can be priced in similar fashion. bility that BT = 0 is 0. (See Harrison and
Kreps (1979) for details.)
Pricing default swaptions To use this result we first let A be a securi-
The growing market in default swaptions has ty that at T pays 0 if default has occurred and
led to a demand for models to price these otherwise pays the upfront cost (as of T) of a
products. First, let us clarify terminology. An zero-premium CDS with the same maturity
option to buy protection is a payer swaption as the CDS underlying the swaption. We let
and an option to sell protection is a receiver B be a security that at T pays 0 if default has
swaption. This terminology is analogous to occurred and otherwise pays PV01T. With
that used for interest rate swaptions. these definitions the ratio AT /BT is equal to
Blacks formulas for interest rate swap- the spread ST if default has not occurred at T,
tions can be modified to price European otherwise AT /BT and ST are not defined. The
default swaptions. Consider a European distribution of ST should then be such that
payer swaption with option expiry date T and E[ST ] = A0/B0 where the probability of
strike spread K. The contract is to enter into default before T is put to 0. A0/B0 is the T-for-
a long protection CDS from time T to TM, and ward spread, denoted F0, for the underlying
it knocks out if default occurs before T (see CDS where the forward contract knocks out
Figure 29). Conditional on surviving to time if default occurs before T.
T, the option payoff is The next step is to let A be the swaption,
in which case AT is 0 if default happens
before T and PST otherwise. The value of the
PST = PV 01T max{ST K ,0} swaption today is then
PS
where the PV01T is the value at T of a risky PS 0 = B0 E T = PV 010 E [max{ST K ,0}]
1bp annuity to time TM or default, and ST is BT
the market spread observed at T on a CDS
with maturity TM. If we make the assumption that log(ST) is
Ignoring the maximum, this is the stan- normally distributed with variance 2T, corre-
dard payoff calculation for a forward start- sponding to the spread following a log-nor-
ing CDS. See page 32 for a discussion of mal process with constant volatility , then
CDS pricing. with the requirement E(ST) = F0 (the forward
From finance theory we know that for any spread), we have determined the distribution
given security, say B, that only makes a pay- of ST to be used to find E[max{ST K,0}]. It is
ment at T, there is a probability distribution easy to calculate this expectation and we
of the spread ST such that for any other arrive at the Black formula
security, say A, that also only makes a pay-
ment at T, the ratio A0/B0 of todays values of
the securities is equal to the expectation of PS 0 = PV 010 (F0 N (d1 ) K N (d 2 ) ),
the ratio AT /BT of the security payments at T. log( F0 / K ) + 2T / 2
d1 = d 2 = d1 T
The result is valid even if BT can be 0 as long T
as AT = 0 when BT = 0. The states where BT
where N is the standard normal distribution swap maturity 20/9/2008. The valuation
function. The Black formula for a receiver date is 28/8/2003. The strike is 260, which
swaption is found analogously. (See Hull and is the five year CDS spread on the valuation
White (2003) and Schonbucher (2003) for date. Our trading desk quoted the swaption
more details.) at 2.29%/2.52%, which means that $10m
It is important that the forward spread, F0, notional can be bought for $252,000 and
and PV010 are values under the knockout sold for $229,000. These are prices of non-
assumption. Calculation of F0 and PV010 knockout options. The fair value of protec-
should be done using a credit curve that has tion until swaption maturity given the credit
been calibrated to the current term structure curve on the valuation date is 0.211%, the
of CDS spreads. PV01 used in the Black formula is 4.027 and
To value a payer swaption that does not the forward spread is 274.7bp. These num-
knock out at default, we must add the value bers imply that the bid/offer volatilities
of credit protection from today until option quoted by our desk are approximately 75%
maturity. Under a non-knockout payer swap- and 85%.
tion, a protection payment is not made until
option maturity. The value of the credit pro- Interest rates and credit risk
tection is therefore less than the upfront One way to model credit spread dynamics is
cost of a zero-premium CDS that matures via a hazard rate process. This provides a
with the option. The knockout feature is not consistent framework for modelling spreads
relevant for receiver swaptions, as they will of many different maturities. The differences
never be exercised after default. between directly modelling the credit
The last step in valuing the swaption is to spread and modelling the hazard rate are
find the volatility to be used in the Black similar to the differences between mod-
formula. An estimate of can be made from elling the yield of a bond and using a term
a time series of CDS spreads. Examination structure model as discussed above.
of CDS spread time series also reveals that A stochastic hazard rate model is natural-
the log-normal spread assumption often is ly combined with a term structure model to
inappropriate. It is not uncommon for CDS produce a unified model that can, at least in
spreads to make relatively large jumps as theory, be used to price both bond options
reaction to firm specific news. However cal- and credit spread options. However, a num-
ibrating a mixed jump and Brownian process ber of practical complications arise in get-
to spread dynamics is not easy. ting such an approach to work. Since bond
To price default swaptions with and the CDS markets have their own
Bermudan exercise we could construct a dynamics, usually not implying the same
lattice for the forward spread, but criticism issuer curves, it may not be appropriate to
analogous to that for using a yield diffusion naively price all options in the same cali-
model to price bond options applies. brated model without properly adjusting for
Instead, we recommend using a stochastic the basis between CDS and bonds.
hazard rate model. Also, calibration of the volatility parame-
To illustrate the use of the Black formula, ters of the hazard rate process, when possi-
consider a payer swaption on Ford Motor ble, is less than straightforward, especially
Credit with option maturity 20/12/2003 and when calibrating to bond options struck on
price or yield, or to bonds with embedded an option to enter into a receiver swap with
options. Interest rate volatility parameters fixed rate k for the remaining life of the
should be calibrated separately using liquid original trade at default. For simplicity, we
prices of interest rate swaptions. These assume that default can only take place at
parameters are then taken as given when times ti. If B denotes the price process of
determining the hazard rate volatility param- the savings account, then computing the
eters by fitting to time series estimates or expected discounted cash flows gives the
calibrating to market prices. value V0 for the price of the default protec-
Finally, it is necessary to determine the cor- tion, where
relation between interest rates and hazard
rates. When estimating correlations it is
n max (RSt , 0 )
important not to use yield spread or OAS
V0 = E i
= ti P [ = ti ]
(option adjusted spread) as a proxy for the i =1 B (ti )
hazard rate, because such spread measures
do not properly take into account the risk of
default. Using OAS is especially a problem for The interpretation of this equation is that the
long maturity bonds with high default proba- value of default protection is a probability
bilities and high recovery rates. For such weighted strip of receiver swaptions, where
bonds there can be a significant decrease in each swaption is priced conditional on
OAS when interest rates increase even if haz- default happening at ti.
ard rates remain unchanged. Correlations Representing the default protection via a
should be estimated directly using bond strip of swaptions is a very useful framework
prices or CDS spreads. for developing intuition around the pricing
and helps us understand the importance of
Modelling hybrids the shape of the interest rate curve. In terms
The behaviour of hybrid credit derivatives is of volatility exposure, the protection seller
driven by the joint evolution of credit has sold swaptions and is therefore short
spreads and other market variables such as interest rate volatility.
interest and exchange rates, which are com- If the rate and credit process are correlat-
monly modelled as diffusion processes. As ed, then there will also be a spread volatility
a consequence, the reduced form approach dependence. However, under the assump-
is the natural framework for pricing and tion of independence, the volatility of the
hedging these products. credit spread does not enter into the valua-
We illustrate the main ideas via the exam- tion, only the default probabilities.
ple of default protection on the MTM of an A strip decomposition, as we have shown
interest rate swap. Suppose an investor above, is the basic building block for most
has entered into a receiver swap with fixed hybrid credit derivatives. A cross-currency
rate k with a credit risky counterparty. If the swap for example could be dealt with in
MTM of the receiver swap, RSt is positive exactly the same way, with the exchange
to the investor at the time of default, this is rate as an additional state variable. For this
paid by the protection seller. If RSt is nega- instrument, the exchange rate exposure is of
tive, the investor receives nothing, so that prime importance.
the payoff at default is max(RSt,0). This is In terms of tractability, it is important to be
able to calculate the conditional expecta- structure between credit spreads and the
tions appearing in the strip decomposition. other market variables is the main challenge
A further consequence of the hazard rate in modelling hybrids. The simplest approach
method is that we do not actually have to is to work within a diffusion setting where
condition on the realisation of the default spreads and interest rates/FX are correlated.
time itself, conditioning instead on the reali- Such a model will not necessarily generate
sation of the hazard rate process. This is par- levels of dependence representative of peri-
ticularly advantageous for Monte Carlo ods of market stress where investment
simulation, as one does not have to explicit- grade defaults are likely. As a starting point,
ly simulate default times and is a useful vari- however, it appears to be reasonable.
ance reduction technique. Even within this framework, the effect of
The parameters needed for pricing correlation on the valuation of a hybrid
hybrids are essentially volatility and depen- instrument can be marked. At this stage of
dence parameters. Calibrating volatilities the market it is safe to say that this correla-
for interest rates and FX is relatively tion is a realised parameter as opposed to
straightforward. Credit spread volatilities an implied one, ie pricing and hedging
are somewhat more involved. Until recent- must proceed on the basis of a view on cor-
ly, we have had to rely on estimates of his- relation founded on historical estimates.
torical volatilities; now the growing options Going forward, it will be interesting to see
markets are starting to make the calibration to what extent a market in implied correla-
of implied spread volatilities feasible. tions will develop via standardised hybrid
Determining the correct dependence credit derivatives.
References
Altman E, Resti A and Sirone A, 2001 Ganapati S, Ha P and OKane D, 2001
Analyzing and explaining default Synthetic CDOs how are they structured?
recovery rates how do they work?
Report submitted to ISDA, Stern School of CDO Monthly Update, Lehman Brothers,
Business, New York University, December November
Berd A and Naldi M, 2002 Ganapati S and Ha P, 2002
New estimation options for the Lehman Evolution of synthetic arbitrage CDOs
Brothers risk model Structured Credit Strategies, Lehman
Quantitative Credit Research Quarterly, Brothers, November
Lehman Brothers, September Ganapati S, Berd A, Ha P and
Berd A and Ranguelova E, 2003 Ranguelova E, 2003
The co-movement of interest rates and The synthetic CDO bid and basis conver-
spreads: implications for credit investors gence which sector is next?
High Grade Research, Lehman Brothers, Structured Credit Strategies, Lehman
Embrechts P, Lindskog F and Brothers, March
McNeal A, 2003 Gregory J and Laurent J-P, 2003
Modelling dependence with copulas and I will survive
applications to risk management Risk June, pages 103107
In Handbook of Heavy Tailed Distributions in Harrison J and Kreps D, 1979
Finance, Elsevier, New York, pages 329384 Martingales and arbitrage in multiperiod
Finger C, 1999 securities markets
Conditional approaches for CreditMetrics Journal of Economic Theory 20, pages
portfolio distributions 381408
RiskMetrics Group, New York Hull J, 2000
Finger C, 2000 Options, futures and other derivatives
A comparison of stochastic default Fourth edition, Prentice Hall
rate models Hull J and White A, 2001
RiskMetrics Group, New York Valuing credit default swaps II: modelling
Fitch Ratings, 2003 default correlations
Default correlation and its effect on Journal of Derivatives 8, pages 1222
portfolios of credit risk Hull J and White A, 2003
Structured Finance, Credit Products Special The valuation of credit default swap options
Report, February Working paper, University of Toronto
Frye J, 2003 Isla L, 2003
A false sense of security Hedged CDO equity strategies
Risk August, pages 9599 Fixed Income Research, Lehman Brothers,
September
Approved by Lehman Brothers International Risk Waters Group and Lehman Brothers and the
(Europe), which is authorised and regulated by the author have made every effort to ensure the
Financial Services Authority. Rules made for the accuracy of the text; however, neither they nor any
protection of investors under the UK Financial company associated with the publication can
Services Act may not apply to investment business accept legal or financial responsibility for
conducted at or from an office outside the UK. consequences, which may arise from errors,
omissions, or any opinions given.
Published by Incisive Risk Waters Group,
Haymarket House, 2829 Haymarket,
London SW1Y 4RX
2003 Lehman Brothers and Incisive RWG Ltd
Notes
An Established Tradition
of Research Excellence
Weekly commentary The Lehman Brothers Global Fixed Income Research
across asset classes Group has been consistently recognised as a top-ranked
research department.
All Rights Reserved. Member SIPC. Lehman Brothers International (Europe) is regulated by the Financial Services Authority. 2003 Lehman Brothers Inc.
lehman cover.qxd 10/10/2003 11:03 Page 1