HVB Group DJ ITRAXX Credit at Its Best

Global Markets Research
European Credit Strategy
Credit Derivatives Special

October 5, 2004
DJ ITRAXX: CREDIT AT ITS BEST!

Official DJ iTraxx A COMPREHENSIVE GUIDE TO THE MOST LIQUID
presentation
INSTRUMENT IN THE CREDIT UNIVERSE
Click here
Following our first Credit Derivatives Special “CDS – Mechanism, Pricing &
Application”, in this issue we focus on the DJ iTraxx family. We include a
detailed description of single instruments, the implementation of trade ideas
and an in-depth analysis of different risks related to DJ iTraxx investments.
Contents To account for the rapid acceleration of new instruments and innovations in
Product overview ___________2 the credit derivatives world, we will update this publication regularly.
Trading Mechanism _________9
Pricing & quotation ________11 n Introduction: The DJ iTraxx index family offers well-diversified exposure to the
DJ iTraxx tranches _________24
Trading strategies__________47 EUR credit universe, including investment grade and crossover names. Besides
Product outlook ___________57 plain-vanilla indices, tranches, options and FTD baskets are available.
Hot topics ________________58
Appendix _________________61 n Pricing and quotation of DJ iTraxx swaps: It seems that an index is equivalent
to a portfolio of standard CDS contracts including all index constituents.
Admittedly, this holds for the default leg but not for the premium leg. While
maturity mismatch and quoting conventions complicate the valuation in theory,
in practice, prices are driven by supply and demand rather than by models.
Authors n DJ iTraxx tranches – valuation, modeling aspects & sensitivities: Tranched
Dr. Jochen Felsenheimer
+49 89 378-18188 DJ iTraxx products are quite similar to synthetic CDO transactions, but with an
Jochen.Felsenheimer@hvb.de important difference: DJ iTraxx tranches are standardized and very liquid.
Modeling highly complex tranche sensitivities is required for an accurate pricing
Dr. Philip Gisdakis
+49 89 378-13228 model, which is crucial for hedging rather than trading activities.
Philip.Gisdakis@hvb.de
IT IS NOT ALL ABOUT SPREAD SWINGS: HIGHLY COMPLEX TRANCHE SENSITIVITIES
Michael Zaiser
+49 89 378-13229 60%
Equity
Michael.Zaiser@hvb.de BBB
present value change (in % of the notional)
40% AAA
Junior
Senior
20%
0%
HVB CDS/DJ iTraxx Trading

Oliver Reisinger -20%
+49 89 378-17600
HVB DJ iTraxx Marketing -40%

Theodor Müller
+49 89 378-18386
-60%
10 20 30 40 50 60 70 80 90 100
mutual spread level (in bp)
Source: HVB Global Markets Research
n Trading the DJ iTraxx: The DJ iTraxx universe offers the opportunity to

Bloomberg implement various trading strategies, including plain-vanilla intra- DJ iTraxx
HVBR
trades, inter-basket trades, correlation arbitrage, cross asset class strategies,
Internet etc. Without claiming completeness, we highlight the most popular trading
www.hvb.de/valuepilot strategies and analyze payoff structures and risk factors.
© HVB Corporates & Markets, Global Markets Research.

See third to last page for disclaimer.
A member of HVB Group
Credit Derivatives Special - DJ iTraxx: Credit at its best!

October 5, 2004
PRODUCT OVERVIEW
A star is born: the iTraxx will The merger between iBoxx Ltd and Trac-x LLc created a new entity, the
change the credit universe
International Index Company (IIC). ICC will include the current iBoxx cash
indices and the credit derivative index business of iBoxx and Trac-x in
Europe, in the US and in Asia. The label for the bond indices is Dow Jones
iBoxx, while the derivatives indices will be known as Dow Jones iTraxx. In
this publication, we focus on credit derivative indices, which provide highly
liquid and transparent trading and investment opportunities within the
European credit market.
The iTraxx family includes high- The iTraxx index family offers exposure to the whole EUR denominated credit
grade and high yield names, …
universe, including investment grade and high yield names. Given the high
liquidity of the index family, investors are able to implement several strategies.
Besides taking directional market risk, the index family offers the opportunities
for hedging activities (on a macro level) and allows relative value plays between
the overall market and single sectors, between different sectors or between
different market segments (subs vs. senior financials, HG versus HY).
… in different sub-indices and In combination with additional released index derivatives (credit spread options
even tranches
and standardized tranches), also highly “specialized” investors could benefit
from the iTraxx. Correlation-, dispersion- and volatility trading activities could
be easily established, supported by the high liquidity and small bid-offer spreads
in these instruments.
INDEX OVERVIEW
Wide range of iTraxx sectors The iTraxx family includes a wide range of sector indices. Moreover, the
available
Benchmark Index (DJ iTraxx Europe) is also the underlying for derivatives
like credit spread options, standardized tranches and future contracts
(forthcoming).
THE ITRAXX FAMILY: A WIDE RANGE OF INSTRUMENTS AVAILABLE*
DJ iTraxx Europe Sub-Indices

(125 most liquid names
in the CDS market) Including the most liquid names
in the sector
- Non-Financials (100)
- Financials Senior (25)
- Financials Sub (25)
DJ iTraxx Europe HiVol - TMT (20)
(30 highest spread - Automobiles (10)
names in the DJ iTraxx) - Industrials (20)
- Energy (20)
- Consumers (30)
DJ iTraxx Corporates
(100 largest issuer Derivatives
from the iBoxx)
- Tranched iTraxx (standardized
tranches based on the DJ iTraxx)
- iTraxx Options (on spread moves
DJ iTraxx Europe of the DJ iTraxx)
Crossover - First-to-default baskets
(30 HY references) - iTraxx Futures (forthcoming)
© HVB Corporates & Markets, Global Markets Research. 2


October 5, 2004
DJ ITRAXX EUROPE
The so-called Master index is The so-called Master index is the DJ iTraxx Europe, which includes the 125
the DJ iTraxx Europe, …
most liquid names in the credit default swap universe. CDS volumes of single
names through a dealer poll define liquidity, while the index composition is
administered by the IIC. The DJ iTraxx Europe can actually not be
characterized as an index due to the fact that no index level is calculated by
the IIC. It is rather a basket of singles names, which acts as underlying for
several swap and tranched products.
… which includes the most Quoted prices for these products are a result of supply and demand within credit
liquid 125 names, …
markets. However, Mark-IT, the leading provider for credit default swap levels,
provides market quotes for the underlying “index” constituents. All 125 names
are equally weighted for simplicity reasons. This means that the DJ iTraxx
Europe is built as a diversified credit index rather than as a tracking instrument
for investors who want to replicate the cash credit market. The index will be
tradable unfunded (CDS format) and funded (note issued by iBond Securities Plc.
and rated by S&P and Moody’s), with standard maturities of 5Y and 10Y for the
swap and 5Y for the note. There will be a new series of the “index” every six
months. The first two series, however, have a different maturity profile to bring
the DJ iTraxx Europe index in line with its counterparts in Asia and the US.
… based on trading volumes The portfolio construction rules are based on following criteria. Each market
maker submits a list of 200 – 250 names, which are European-listed and have
the highest trading volume over the previous six months, excluding internal
transactions. The volumes of names which have the same “corp ticker” on
Bloomberg but trade separately in the CDS market (e.g. Arcelor and Arcelor
Finance), are summed up to receive the overall issuer volume. Thus, all market
makers send a list ranked by issuer volume to the index provider, who excludes
all Baa3/BBB- rated names with a negative outlook. Each issuer is allocated to
an appropriate DJ sector and then mapped to an iTraxx sector following a
sector-mapping schedule. The ranking within one sector is also based on the
trading volume, while the sector weighting (please refer to the overview above)
is not directly linked to the ranking of single issuers. This means the index
includes 20 TMT issuers, even in case the 100 names with the highest trading
volume would be within the sector. Despite trading volumes, all constituents
within the overall index and single sectors are equally weighted.
The master index is the most The master index is the most liquid product within the index family, which
liquid instrument within the
iTraxx family
provides a broad-based exposure to European credits/entities. The index does
not include non-European names, with GE, GM and Ford being the most popular
exclusions. Given its well-diversified character, the DJ iTraxx Europe could be
used to implement basic core-satellite strategies, with the index being the basic
exposure to credits. With the help of single-name CDS and sector sub-indices, a
strategic sector view as well as a fundamental bottom-up view could be
implemented. Moreover, the index will be the leading instrument for portfolio
hedges, not only for credit portfolio managers, but also for trading desks and
treasury activities. As the DJ iTraxx Europe acts as the underlying for
standardized tranches and options/futures, also correlation traders will use the
index to implement trade ideas and hedge for directional (market) risks.


October 5, 2004
DJ ITRAXX EUROPE HIVOL INDEX

DJ iTraxx Europe HiVol offers The DJ iTraxx Europe HiVol index contains the 30 (equally-weighted) names
access to the 30 names with
the widest CDS spread
from the DJ iTraxx Europe with the widest 5Y credit default swap levels
without any sector limitations. The index is only tradable in swap format
(unfunded), with standard maturities being 5Y and 10Y, while a new series
will be issued every six months.
Regarding the risk & return profile, the index is rather an instrument for hedge
funds and prop desks than for portfolio managers. In our view, the construction
of the index bears the risk of a high exposure skewed to single sectors. The 30
entities which carry the widest spread, could be found in subordinated rather
than senior financials, in autos rather than in energy and in industrials rather
than in TMTs. That said, sector-specific risk is clearly dominating pure market
directional risk.
DJ ITRAXX EUROPE CORPORATE

DJ iTraxx Europe Corporate is The DJ iTraxx Europe Corporate is separated from the DJ iTraxx Europe as
based on the 100 most liquid
names out of the iBoxx index,
it also includes non-European entities issuing EUR bonds, with GM and Ford
… being the most popular ones. The index contains the 100 largest non-
financials issuers from the DJ iBoxx EUR Corporate Overall index, based on
cash bonds.
All constituents are weighted with respect to their market capitalization and the
duration of their single issues within the iBoxx Corporates Index. The DJ iTraxx
Europe Corporate index is tradable unfunded and funded with a standard
maturity of 5Y. The note is rated by S&P and Moody’s. A new series will be
issued every 6 months and the index is administered by IIC. The index
construction should fit the needs of asset managers, allowing to track the
performance of the cash iBoxx corporate universe.
… weighted by its duration The portfolio construction follows, in many points, the philosophy of the DJ
value
iTraxx Europe, while the weightings are duration-adjusted. Using the iBoxx EUR
Corporate Overall index (available on iBoxx.com or Deutsche-Boerse.com), the
index provider calculates the duration value of all issuers within the index,
aggregating the market value of all single bonds multiplied by the annual
modified duration. Each issuer within the index will be weighted by its duration
value, ignoring financial reference entities.
Duration-weightings must be The duration-linked constituents weighting is superior to simply equal-weighted
taken into account by
calculating hedge ratios
indices, but does not optimally satisfy the needs for pure spread investors.
“Duration-weighted” means that the market cap of single bonds is adjusted by
its duration. This may have significant impact on the weighting of single names.
The most obvious case is Ford, which has a short average duration and
therefore, lost importance in case of duration-adjusted weightings. However,
this is only half the truth as duration weightings would be appropriate for
portfolio managers who manage their duration exposure without using FI
futures. For pure spread investors, duration-weighted indices do not adequately
reflect the risk positions in their cash credit portfolio. This produces a tracking
error, which must be taken into account when calculating hedge ratios.
However, a majority of fund managers is using duration-adjusted weightings to
analyze their overall risk exposure to single names. For those investors, the
index construction makes sense.


October 5, 2004
DJ ITRAXX EUROPE CROSSOVER

DJ iTraxx Europe Crossover The DJ iTraxx Europe Crossover is separated from the above mentioned
includes 30 names …
universe, as it contains only so-called “crossover” names. The index includes
30 equally weighted entities, whose ratings are not better than Baa3/BBB-
with a negative outlook.
… which carry no better-rating Similar to the DJ iTraxx Europe, construction rules are based on a dealer poll
than Baa3/BBB- with neg.
outlook
and the index is administered by IIC. The index is tradable in CDS format and as
a credit linked note (in contrast to the indices above, rated by S&P only) with
standard maturities of 5Y and 10Y for the swaps and 5Y for the note. The DJ
iTraxx Europe Crossover index provides liquid access to lower-quality names,
which offer a higher expected return to the disadvantage of a higher volatility
(compared to both the DJ iTraxx Europe and to the DJ iTraxx Europe Corporate
indices. That said, this index is likely to be used by hedge funds and prop desks
rather than by credit portfolio managers.
Additional construction criteria The portfolio construction encompasses additional criteria. Each market maker
proposes at least 30 reference entities, which are (a) European non-financials,
(b) have no better rating than Baa3/BBB- with a negative outlook, (c) carry a
minimum spread of twice the DJ iTraxx Europe non-financial spread and a
maximum spread of 1,250 bp or a max. up-front payment of 35% and (d) the
outstanding volume of publicly traded bonds must exceed EUR 100 mn. After
eliminating guaranteed affiliates and wholly-owned subsidiaries of an already
included entity, the 30 entities which receives the greatest number of dealer
votes will be included in the index. Following the poll, each market maker
submits a fixed rate for the index (in spread terms). The average, calculated
from these spreads and rounded up to the nearest 5 bp level, will be the
premium.
SPREAD OPTIONS
Spread options could be used There are also spread options available, which are not standardized and
for hedging purposes and
more complex trading
hence, liquidity should be rather low. However, options offer the opportunity
strategies, … of more complex hedging and trading strategies, which generate interesting
payoff structures. An exact definition of an investor’s risk positions (the
Greeks: delta, gamma, vega and theta), requires an appropriate pricing
model.
A possible approach is to use the Black & Scholes framework, extended by a
stochastic jump process to cover the problem that credit spread swings are not
log-normally distributed. While the Black & Scholes approach is widely
accepted, the problem lies, in our view, in determining the parameters for the
… but the investor is exposed jump process. “Jump intensity” and “jump width”, the two major parameters for
to so-called “model risk” a jump process vary significantly over time. Estimating jump processes depends
heavily on the market environment. As long as there is no standard model to
price credit spread options, investors are exposed to so-called model-risk. An
investor may calculate the theoretically fair price of the option correctly, but all
other market players use different models and calculate diverging prices. In this
case, P&L realization would be affected by the model used by the investor.


October 5, 2004
DJ TRANCHES
Standardized tranches offer Based on the DJ iTraxx Europe 5Y, the DJ iTraxx family even comprises a
leveraged credit exposure
series of tradable and standardized tranches, which offer leveraged
exposure to the European high-grade credit market. Besides long positions,
shorting tranches is also possible.
STRUCTURE OF STANDARDIZED ITRAXX TRANCHES
DJ iTraxx Europe 5Y - Reference Pool
Cash flow
Losses
Equity* BBB* AAA*

(0-3% / 3% / 0%) (3-6% / 3% / 3%) (6-9% / 3% / 6%)
Junior Super Junior Super

Senior High* Senior Low*
(12-22% / 10% / 12%) (9-12% / 3% / 9%)
Investment grade* (3-100% / 97% / 3%)
*in brackets: (tranches / thickness / subordination)

Despite synthetic CDO The tranches are split up to 0-3% (equity), 3-6% (BBB), 6-9% (AAA), 9-12%
transactions, iTraxx tranches
are highly liquid
(junior super senior low), 12-22% (junior super senior high) and 3-100%
(investment grade), with the percentage related to total loss. This means that in
case 10% of all index constituents have a credit event and a recovery value of
80%, the “first-loss-piece” is still intact as total loss is only 2%. A new series will
be launched every six months, while future attachment points could change in
line with the risk profile of the underlying index. From a risk and return profile,
DJ iTraxx tranches equal other investment grade CDOs. The most important
difference is high liquidity in iTraxx tranche products, while synthetic CDOs are
rather a buy and hold investment. While all tranches above the equity piece are
quoted in standard spread terms, the price quotation for the 0-3% tranche is
quite different. A DJ iTraxx equity tranche always pays a constant spread of 500
bp on the (remaining) notional amount while market makers quote a percentage
of the notional amount that has to be paid as an up-front payment when
entering into such a contract as a protection buyer. As tranche pricing is highly
complex and directly linked to several risk factors (directional market moves,
implied correlation, recovery assumptions, spread volatility, etc.), iTraxx
tranches offer an attractive opportunity for pure credit derivative players rather
than for credit portfolio managers.


October 5, 2004
DJ ITRAXX EUROPE FIRST TO DEFAULT BASKETS

Standardized DJ iTraxx FTDs … Since the introduction of the DJ iTraxx Series 2 on September 20,
standardized First-to-default baskets (FTDs) on several DJ iTraxx indices
are tradable. Besides an interesting payoff-profile of an isolated FTD
position, this offers attractive trading opportunities (correlation arbitrage)
in combination with other DJ iTraxx instruments (subindices & tranches).
… are available on the overall Standardized DJ iTraxx FTD’s for each sector are constructed by excluding the
index and on single sectors
two sector constituents with the greatest spread and the two with the smallest
spread, while the five next most liquid entities constitute the basket. Baskets are
available for automobiles, consumers, energy, financials senior, financials
subordinated, industrials, TMT, diversified and HiVol. Constituents of the FTD
baskets will only change if the entity is no longer in the relevant DJ iTraxx index
sector, or if the entity is one of the two names with the greatest or smallest
spreads in its basket. Thus, an entity will be replaced by the next most liquid
eligible name.
The protection seller gained In case of a credit event, the protection seller of an FTD pays 1-recovery value of
leveraged exposure to a basket
of credits
the defaulted issue to the protection buyer. Following the first default, the FTD-
contract is terminated and the protection buyer loses any further protection for
the other basket constituents. Driven by default correlation, the basket spread
lies between the spread of the most risky constituent (which carries the highest
spread) and the sum of the CDS spreads of all basket credits. The protection
seller gains leveraged exposure to a basket of credits as is the case with tranche
investments.
Buying protection on a FTD is Given the sector-wise nature of DJ iTraxx FTDs, combining an FTD position with
only a incomplete hedge for a
long position in the underlying
a plain vanilla sub-index position is an obvious trading strategy. From a hedging
sector index … perspective, buying protection for a FTD is a (incomplete) hedge for a long
position in the underlying sector index. In case of a credit event, the long index
position experiences a loss, while the short basket position generates profit. The
optimal hedge ratio against a default (assuming the same recovery rate for all
sector constituents) using the default leg equals
(1 / n) ⋅ R
basket position = ⋅ index position
1− R
… as it does not account for with n being the number of index constituents and R being the recovery rate.
the risk that another index
member could default
This hedge ratio is far away from being perfect as
– it does not account for the risk that another index member could default,
which is not included in the basket and
– it does not incorporate the premium leg (i.e. payments in the aftermath of a
default on the index, while the basket position is terminated).
Besides the opportunity for correlation arbitrage (mispricing of correlation
instruments like tranches and/or baskets), a combination of long/short basket
positions to actively trade spread and default correlation as well as taking
directional spread risks are interesting alternatives for credit investors.


October 5, 2004
FUTURE CONTRACTS
Futures will be introduced in Future contracts based on the DJ iTraxx Europe will likely be introduced in
early 2005
early 2005. Although the final construction has not been officially
announced yet, the goal is to create a highly liquid derivative instrument
which offers the opportunity to build up / to reduce exposure to credit risk
in an efficient way. The design of the contract (expiry dates, price quotation,
etc.) will be closely linked to Eurex standard futures, like the
Schatz/Bund/Bobl/Buxl contracts.


October 5, 2004
TRADING THE ITRAXX
FUNDED VERSUS UNFUNDED AND THE P&L IMPACT

The swap format offers higher The iTraxx family offers the opportunity to build up unfunded (swaps) and
liquidity
funded (via a note) credit exposure, with the former proving a high liquidity.
In case of a credit event, unfunded instruments are handled by physical
settlement while cash settlement is applied for the note. Every six months, a
new series will be launched, making a roll necessary for investors who want
to stick with the maturity profile of their credit exposure.
Funded iTraxx exposure via There are two ways to create exposure to the iTraxx index family: funded and
iTraxx notes
unfunded. Funded means an investor is buying a credit-linked note, which is
paying Euribor plus the index spread (in the table below, we show index
spreads for all iTraxx notes). Practically, an investor pays 100 (par) for the note
and receives Euribor plus spread on a quarterly basis. However, there are two
payoff legs, the non-credit-event case and the credit-event case. The non-credit
event case is straightforward, as the coupon is paid until maturity, when the
note will redeem at par. In case of a credit event on a reference entity, the
weighting of the entity and the recovery value determine the cash payment.
Provided a weight of 0.8% (DJ iTraxx Europe) and a recovery value of 40%
(exception: 20% for financials sub), the issuer pays the investor the cash
settlement amount of 0.8% × notional × 40%. In case of a notional of EUR 5 mn,
the investor receives EUR 16,000 in cash. Simultaneously, the redemption
amount will be reduced by 0.8% from 100 to 99.2% (corresponding to EUR 4,96
mn in our example). In the aftermath of the credit event, the investor receives
coupon payments on the reduced notional until maturity or until another credit
event occurs. The spread, however, on the note will remain unaffected by the
occurrence of a credit event.
DJ ITRAXX SWAP FAMILY: SERIES 1 AND 2 COUPONS
DJ iTraxx index Maturity Coupon – Series 1 Coupon – Series 2

Europe 5Y 45 bp 35 bp
10Y 60 bp 50 bp
Non-financials / Non-financial sectors 5Y 45 bp 35 bp
10Y 60 bp 50 bp
Financials Senior 5Y 25 bp 15 bp
10Y 35 bp 20 bp
Financials Sub 5Y 45 bp 35 bp
10Y 60 bp 50 bp
Corporate 5Y 50 bp 45 bp
Crossover 5Y 300 bp 275 bp
10Y 315 bp 295 bp
HiVol 5Y 75 bp 55 bp
10Y 90 bp 75 bp
Unfunded iTraxx exposure in An unfunded credit investment through a swap contract is the more liquid
CDS format
alternative to purchasing a note. Such a basket CDS product references to the
issue spread level of the current series (called premium level), e.g. 35 bp for the
DJ iTraxx Europe 5Y (series 2) contract. Keeping the spread (swaps) or coupon
(notes) levels constant over time is simply for the sake of standardization and
hedging. A long position implemented one month ago could perfectly be hedged


October 5, 2004
with a short position today. To keep the premium constant, an investor

receives/pays up-front the amount, which makes the deal PV-neutral.
Apparently, this procedure is a main difference to regular CDS contracts. In case
of a spread tightening following the introduction, an investor who would like to
sell protection needs to pay up-front an amount, which offsets the experienced
spread tightening as the premium, which the investor receives, remains
constant. In the section below, we point out the pricing mechanism for the swap
format.
The P&L of an DJ iTraxx The P&L impact of a movement of the underlying quoted spread could be easily
investment can easily be
calculated using the CDSW
calculated with the CDSW function in Bloomberg, which offers a simple analysis
function on Bloomberg of the payoff structure of iTraxx swap investments (by the way: this is the same
function applicable for single name CDS). We will elaborate on this procedure in
the “Quotation and Valuation for DJ iTraxx Swaps” section.
In case of a credit event, For the moment, let us assume an investor selling protection to a market maker.
physical delivery is stipulated
In case no credit event occurs, the market maker pays the deal spread on the
notional amount to the counterparty (seller of protection) on a quarterly basis.
The counterparty would receive a constant premium until maturity. Physical
settlement is stipulated for the case of a credit event. Assuming such a case for a
reference entity with a weight of 0.8% (1/125), the investor (protection seller)
delivers 0.008 multiplied by the notional amount in deliverable obligations of the
reference entity to the market maker. In the aftermath, the notional amount
reduces by 0.8% points and hence, the market maker pays the constant
premium level for only 99.2% of the original notional until maturity or until any
further credit event occurs. While the continuing premium payment is not linked
to the specific reference entity, which was hit by a credit event, the overall P&L
of the protection seller (and to a lesser extent to the protection buyer) is affected
through the uncertainty of the recovery value. Assuming that an investor, who is
selling protection, does not carry deliverable obligations of all 125 names within
the basket, he is exposed to recovery risk as the expected recovery value
determines the price of the obligations, which the investor has to purchase. The
security specific P&L impact becoming apparent by comparing future cash flows
of an investment in the basket with an investment in all single constituents,
which we provide in the following section.
Buy & hold investors prefer a Trading-oriented investors clearly favor the unfunded format of DJ iTraxx
funded investment, while
trading accounts favor
indices, which are more liquid than their funded counterparts. These players
unfunded exposure trigger short-term spread trends and roll their exposure in case of the
introduction of a new series. Also, short-term oriented hedging activities are
concentrated on the CDS format. From a balance sheet perspective, reporting
issues (economic versus reported positions) play a role. In case of unfunded
investments, a marked-to-market investor has to realize gains or losses if he
wants to switch in the more liquid (new) series.
Roll-over mechanism The first rollover in the DJ iTraxx Europe indices, which took place September
20, shows that the majority of real money accounts is skewed to long-protection.
Besides the maturity extension, this positioning caused slightly wider spreads in
the aftermath of the rollover as long-protection investors sold protection in the
old series (result: tighter spreads in the old series) and bought protection in the
new series (result: wider spreads in the new series).


October 5, 2004
PRICING AND QUOTATION OF DJ ITRAXX SWAPS

Pricing of CDS indices is not as The International Index Company promises a high degree of “transparency”,
easy as it seems at first glance
a criteria that also should encompass the pricing and valuation topics
concerning the DJ iTraxx CDS Indices. At first glance, it seems that such an
index is equivalent to a portfolio of standard CDS contracts including all
constituents. Admittedly, this holds for the default leg, but not for the
premium leg, as will be shown in the following paragraphs. Beyond this
aspect, there are several other issues like maturity mismatch and quoting
conventions, which complicate the valuation of CDS indices in practice.
However, the high liquidity within the DJ iTraxx universe leads to a price
discovery, which is driven by supply and demand rather than by modeling
aspects. In our view, the added value of the following approach is that it
facilitates to identify trading signals in case the market does not
corresponds to the theoretical fair value. This is similar to analyzing the
basis in the cash bond versus CDS framework.
ISSUE NO. 1 - THE “DISPERSION BIAS”

Remember the mechanism how Just like regular credit default swaps on single names, every CDS contract
standardized CDS indices are
set up
on one of the DJ iTraxx Indices comprises two legs, the default leg and the
premium leg. A closer examination of the default leg in case of occuring
credit events shows, that a portfolio of default legs on the constituent
entities (each of equal size and physical settlement stipulated) replicates the
pay-off structure of this leg in a perfect way.
In order to obtain such a duplication portfolio, one usually has to enter into a
number of regular CDS contracts. Each of these credit default swaps includes a
premium leg, whose spread level differs according to the name. Summing up
these premium legs to the premium leg of the replication portfolio reveals the
discrepancy to the premium leg of the CDS index contract in case any default
occurs.
Let’s consider a simple Let’s take a look at a simple numerical example in order to clarify this
example, ...
consideration. Assume an index that consists of just two names (A and B) with
equal weights (50% each). Both titles are actively quoted in the market for single
CDS contracts trading at spreads of 200 bp and 300 bp, respectively. At this
point, suppose the CDS index contract offers a spread level of 250 bp (that is still
to be discussed!). Different scenarios are examined in the following table:
THE IMPACT OF CREDIT EVENTS ON THE PREMIUM LEG OF A CDS INDEX PRODUCT
VS. THE RESPECTIVE REPLICATION PORTFOLIO
no default after A defaults after B defaults

CDS index product pays ... 250 bp 125 bp 125 bp
CDS on A pays ... 200 bp - 200 bp
CDS on B pays ... 300 bp 300 bp -
Replication portfolio pays ... 250 bp 150 bp 100 bp
Difference between CDS index ±0 bp -25 bp +25 bp
and replication portfolio
... showing a mismatch As long as no default occurs, the premium payment of the CDS index product
between the premium leg of a
CDS index product and the
and the respective replication portfolio matches. If A defaults, the notional


October 5, 2004
respective replication portfolio, amount of the CDS index on which premium is paid is reduced by 50%. Thus,
...
future premium payments amount to just 125 bp with regard to the initial
volume of the contract. In contrary, the replication portfolio pays 150 bp owing
to the outstanding CDS contract on B, still paying 300 bp on half of the total
notional amount of the portfolio. Future premium payments of the replication
portfolio, therefore, exceed premium payments of the CDS index product. If B
defaults instead, the impact will be the other way around. The rationale for the
observed mismatch is that reduction of premium payments of a portfolio
comprising standard credit default swaps is based on the individual spread level
at initiation, while premium cash flows of CDS index products are
proportionately reduced according to the name’s weighting, leaving the spread
level constant.
... also raising the question In the example above, we assumed a spread level for the CDS index product
about the adequate spread
level for the CDS index product
(250 bp) that equals the average of market spreads for the underlying names.
But is that conjecture reasonable? As long as no default events are taken into
account, this approach seems self-evident. However, both default scenarios
show a divergence in premium levels (+25bp vs. –25bp). But which one is more
likely to occur? Assuming the same recovery rate for both constituents in case of
default, B clearly has a higher (implied) probability of default due to a higher
spread level. From the perspective of a protection seller who receives premiums,
the CDS index product with a spread of 250 bp seems to be favorable compared
to the replication portfolio. A fair spread of the CDS index product would
therefore be expected to lie beneath the average of the market spreads for each
of the constituents. But by how much?
A SIMPLE PRICING MODEL FOR BASKET CDS

A simple model will help To answer this question, we need a more comprehensive view on how to
analyze the dispersion bias
price single-name and basket CDS contracts. For this purpose, we will
discuss a quite simple model of the CDS market incorporating all substantial
aspects. This will help us understand the pricing relationship between
single-name CDS and CDS index products without being bothered by side
issues.
Let’s assume a discrete one-period model (1 year) with two points in time (t = 0,
1). Credit events may only occur just ahead of t = 1. Premium payments of any
CDS contract (only maturities of 1Y are possible) are made on a yearly basis and
paid in arrears, but only if no default in the underlying title occurred. The risk-
free interest rate for the sole period is denoted by r, calculated on a yearly
compounding convention. Considering a credit default swap with notional
amount N on the name X with a spread of sX, the present value of both legs is
obtained as follows:
– Premium leg:
1 − pX
PLX , 0 = ⋅ sX ⋅ N
1+ r
with p X = (implied) probability of a default within one year for name X
– Default leg:
⋅ (1 − R X ) ⋅ N
pX
DLX , 0 =
1+ r


October 5, 2004
with R X = (expected) recovery rate for name X
At initiation of a CDS contract in t = 0, the spread fixing is governed by the

requirement that the contract should have a PV of zero:
1 − pX
⋅ s X ⋅ N − X ⋅ (1 − R X ) ⋅ N = 0
p !
PV X , 0 = PLX , 0 − DLX , 0 =
1+ r 1+ r
As the nominal N and the risk-free discounting only appear as scaling factors
within the present value formula, the following pricing relationship results:
(1 − p X ) ⋅ s X = p X ⋅ (1 − R X ) or sX =
pX
⋅ (1 − R X ) (1)
1 − pX
Thus, the CDS spread level sX is solely determined by the (implied) probability of
default and the (expected) recovery rate. In addition, the last equation shows the
well-known credit triangle relationship in a slightly modified form.
The adequate spread is still an The introduced framework can now be used to extract the adequate spread level
average of spreads, but a
weighted one
of the CDS index contract. Our index still contains two names, A and B, that are
equally weighted (same notional amount). The observable single-name CDS
spreads sA and sB are determined according to equation (1). We already pointed
out that the default leg of a CDS index contract equals the sum of the default legs
of the constituents, according to their weighting in the index. However, the
premium leg has the special feature of a unique spread (denoted s) for all
components. Given no arbitrage opportunities, the following equation must hold
at initiation (PV of premium leg and default leg must coincide):
1 − pA N 1 − pB
⋅ s ⋅ = A ⋅ (1 − R A ) ⋅ + B ⋅ (1 − RB ) ⋅
N p N p N
⋅s⋅ +
1+ r 2 1+ r 2 1+ r 2 1+ r 2
This yields:
s ⋅ [(1 − p A ) + (1 − p B )] = p A ⋅ (1 − R A ) + p B ⋅ (1 − RB )
After substituting the pricing relationship according to the equation (1) on the
right side, we derive:
s ⋅ [q A + q B ] = s A ⋅ q A + s B ⋅ q B
with q X = 1 − p X = (implied) survival probability for name X (X = A, B )
To tie up loose ends, we finally get a surprising result:
qA qB
s = sA ⋅ + sB ⋅ (2)
q A + qB q A + qB
What do we learn from this? Firstly, the spread of a CDS index contract can be
calculated as an average of individual spreads. Secondly, the calculation is done
by applying a weighted average instead of taking a simple unweighted mean of
spreads. Thirdly, survival probabilities act as weightings in this calculation.
These survival probabilities can be obtained by means of observable CDS
spreads and allegation of recovery rates for the respective names.


October 5, 2004
WEIGHTED VERSUS UNWEIGHTED AVERAGE

A few numerical examples From a technical point of view, there is a difference between applying a
underpin the impact on CDS
index spreads
weighted vs. an unweighted average. But do these differences lead to
significant discrepancies in practice? As mentioned before, one is firstly
required to extract survival probabilities from observed spreads.
This is done with the following formula which can be obtained by basic
conversions of equation (1):
1 − RX
qX =
1 − RX + s X
According to our numerical example (sA = 200 bp, sB = 300 bp) and the allegation
of a uniform recovery rate of 40% (RA = RB = 0.4), we get survival probabilities of
qA = 0.967742 and qB = 0.952381. Applying equation (2) yields a spread level of
249.6 bp. As expected, the obtained result lies beneath the unweighted average
of 250 bp, but the deviation from this focal point is rather limited. From a
theoretical point of view, the spread level of a CDS index product may also lie
above the unweighted average! Presuming different recovery rates for the
constituent names A and B (such as RA = 0.7 and RB = 0) results qA = 0.9375 and
qB = 0.970874. Although A has a lower spread than B, the survival probability
for A is less than for B. In this case, the spread level of our CDS index contract
should be 250.9 bp. Needless to say that this is an unrealistic setting for DJ
iTraxx CDS Indices.
The discrepancy could be quite As a result, the discrepancy between an unweighted and a weighted average
b igger in a realistic setting
based on (implied) survival probabilities seems to be fairly small in practice. But
it has to be taken into consideration that we just worked with a very simple one-
period model, while the DJ iTraxx Indices (unfunded products) offer terms of 5
and 10 years. Contracts with longer terms should bear lower survival
probabilities, and therefore, a higher dispersion of these probabilities.
Consequently, we would expect the discrepancy of the “averaging effect” to be
higher. Hence, we should put some more effort into this topic, now with a
realistic setting. Focusing just on the premium leg of CDS contracts will do, as
there is no difference between the default leg of a CDS index product and a
basket of single-name CDS contracts. Imagine a CDS contract on the name X,
paying a constant premium (spread sX) at regular dates. These dates are denoted
by t1, t2, ..., tn with 0 = t0 < t1 < t2 < ... < tn = T (maturity). The risk-free term
structure of interest rates is represented by a set of discounting factors
d(ti) (i = 1, 2, ..., n), that denote the present value of a unity payment in ti. Given
this notation, the present value of the premium leg at time t = 0 is represented
by
n
PLX = s X ⋅ ∑ (ti − ti −1 ) ⋅ d (t i ) ⋅ q X (ti )
i =1
with q X (ti ) = probability for name X to survive time ti without default
Stipulating the survival probability function qX(t) is the core challenge in this
equation. The specification of a spread curve and a recovery rate for name X
and the choice of a CDS valuation model (e.g. the JPMorgan model or the
modified Hull-White model) substantially determine the level and shape of that
function and moreover the present value of the premium leg.


October 5, 2004
THE FAIR SPREAD OF A CDS INDEX

The fair spread of a CDS index Let’s now suppose an index comprising equally weighted names j = 1, 2, ..., m.
product remains a weighted
average, but a more complex
The CDS index product should pay a uniform spread s, so that the present
one values of (1) the premium leg from this product and (2) a basket of generic CDS
contracts on the index constituents are the same:
m
 n
 m  n

∑ s ⋅ ∑ (t
j =1 
− ti−1 ) ⋅ d (ti ) ⋅ q j (ti ) = ∑  s j ⋅ ∑ (ti − ti −1 ) ⋅ d (t i ) ⋅ q j (ti )
i
1 44i4
=1
442444443 1 j =1 
444 i =1
4442444444 3
premium leg PV for the index product sum of premium leg PV's for single name CDS
Solving for s results in:
 n
  n

m  ∑ (t i − t i −1 ) ⋅ d (t i ) ⋅ q j (t i )  m  ∑ d (ti ) ⋅ q j (t i ) 
s = ∑  s j ⋅ mi=1 n  ≈ ∑  s j ⋅ mi=1 n .
j =1   j =1  
 ∑∑ ( t i − t i −1 ) ⋅ d ( t i ) ⋅ q (
k i t )  ∑∑ d ( t i ) ⋅ q ( t
k i )
 k =1 i =1   k =1 i =1 
The spread of the CDS index product is still a weighted average of single-name
CDS spreads. Assuming equally spaced premium payments, which is almost
correct in practice, these weightings are substantially determined by survival
probabilities qj(tI), while discounting factors d(ti) only play a minor role.
Weightings are referred to as The term applied for weighting individual spread levels, namely
“risky duration” or “risk basis
points” n
∑ d (t ) ⋅ q (t ) ,
i =1
i j i
is often referred to as “risky duration” or “risky basis points” (“risky DV01”). It

represents the marginal change in present value of the respective single name’s
premium leg with respect to a marginal change in the individual spread level sj.
SIMULATING THE DISPERSION BIAS

The spread bias for the DJ As it is fairly hard to understand the impact of the above formula on the
iTraxx Benchmark Index is
rather limited ...
pricing of CDS index products, we decided to perform a simulation for the
DJ iTraxx Benchmark Index and its subindices.
The table below shows the results based on observed single-name CDS spreads
from August 11 and the risk-free interest rate curve at that time (from
Bloomberg’s CDSW function). For simplicity reasons, we assumed flat spread
curves and a constant recovery rate of 40%, which enables us to derive the
survival probabilities easily with the help of the famous credit triangle.
... until there is no spread The dispersion bias, measured as the difference between the unweighted
blowout of a single-name
average of single-name CDS spreads and the weighted average in accordance
with the formula above, seems to be rather limited. The highest bias (0.72 bp) is
obtained for the 10Y “Consumer Cyclical” subindex. As all calculated biases in
the table above are below 1 bp, the problem under discussion seems a bit
exaggerated. That is not totally true when you think of a setting in which a
single-name is close to default accompanied by a spread blowout of the
respective name. Let’s assume a spread widening of an entity within the “Auto &
auto parts” subindex to a new level of 500 bp, retaining the spreads of all other
constituents. Our simulation then yields a spread bias of 11.55 bp, which is
quite considerable when pricing such an index product!


October 5, 2004
THE DISPERSION BIAS FOR THE DJ ITRAXX BENCHMARK AND ITS SUBINDICES
5Y contracts 5Y contracts 10Y contracts 10Y contracts

spread bias std.deviation spread bias std.deviation
DJ iTraxx Benchmark (125) 0.17 bp 20.1 0.48 bp 24.7
Non-financial corporates (100) 0.15 bp 19.1 0.41 bp 22.9
Auto & auto parts (10) 0.17 bp 20.1 0.55 bp 26.6
Consumers (30) 0.25 bp 24.4 0.62 bp 28.3
Energy (20) 0.04 bp 9.5 0.12 bp 12.5
Industrials (20) 0.09 bp 14.5 0.22 bp 16.6
TMTs (20) 0.03 bp 9.0 0.11 bp 11.8
Senior Financials (25) 0.02 bp 6.0 0.04 bp 7.4
Subordinated Financials (25) 0.04 bp 9.6 0.15 bp 13.6
Source: HVB Global Markets Research, Mark-It, and Bloomberg. The columns named “std.deviation” denotes the
standard deviation of single-name CDS spread levels of respective index constituents at a point of time.
The spread bias is primarily Columns 3 (for 5Y contracts) and 5 (for 10Y contracts) offer another interesting
attributable to spread
dispersion
insight. There, we show the standard deviation of the respective single-name
CDS spreads, which is also measured in bp. The left scatter plot below depicts
that the spread bias (columns 2 and 4) is substantially explained by the spread
variance (which is the square of the standard deviation) given the term of the
contract. Strictly speaking, the relationship between spread variance and the
spread bias can suitably be approximated through a linear function.
THE DEPENDENCY OF THE SPREAD BIAS ON THE DISPERSION OF SPREADS

0.8 40
5Y Indices 5Y term
0.7 35
standard deviation of CDS spreads
10Y Indices Consumers 10Y term

0.6 Linear (10Y Indices) 30
sp Linear (5Y Indices) Auto & auto parts
re 0.5 25
ad
bi 0.4 20
as effect of a longer term
in 0.3 15
bp
0.2 10
0.1 5
0.0 0
0 200 400 600 800 1000 0 20 40 60 80 100
2
spread variance within an index (a subindex, resp.) in average CDS spread level
bp
Provided a fixed spread Another important observation can be obtained by comparing the two added
dispersion, a longer term
contract is associated with a
trend lines for 5Y and 10Y indices. A longer term for a CDS index contract is
bigger spread bias associated with a higher dispersion bias given an unchanged spread dispersion.
This seems clear, as survival probabilities qj(ti) have a wider variation for later
dates ti. In addition, having a position in a 10Y contract instead of a 5Y contract
is usually associated with a higher spread dispersion. This is due to the fact that
individual spread curves are increasing functions in maturity accompanied by a
higher standard deviation of spreads for longer terms. The scatter plot above on
the right underpins this relationship, where we show the dependency of spread
dispersion on the average level of spreads within an index or subindex,
respectively. It is not surprising that 10Y contracts on Consumers and Autos
maintain an outlying position in the left chart above.


October 5, 2004
ISSUE NO. 2 - THE “MATURITY MISMATCH”

Maturity mismatch may be Applying an unweighted average of single-name CDS spreads for pricing DJ
another reason for deviations
iTraxx swap products may be a rough approximation, except for a few
extreme situations mentioned above. However, there is another possible
mismatch that has to be taken into consideration.
Different rollover periods in Ordinary CDS contracts have a rollover period of 3 months (owing to standard
single name CDS and the DJ
iTraxx universe
payment dates March 20, June 20, September 20, and December 20), while new
DJ iTraxx Indices are only set up every six months (March 20, September 20).
During the first three months of an index contract’s life, maturity fits to those of
standard single-name contracts quoted in the CDS market. After this period of
time, there is a lack of appropriate CDS quotations owing to a maturity
mismatch. The due date of quoted 5Y and 10Y credit default swaps will then
exceed the maturity of respective index products. On December 20, 2004, we
will encounter this phenomenon for the first time in DJ iTraxx history as the
first two series are exceptionally rolled in a three months interval (initiation on
June 21 and September 20, 2004). As the maturity of CDS index products is
reduced by three months in comparison to regularly quoted CDS contracts, we
expect a relative drop of quoted spreads for DJ iTraxx swaps from one day to
another in accordance with the steepness of credit spread curves. For example,
we expect the 10Y DJ iTraxx Benchmark spread to drop about 0.7 bp relatively
to quoted 10Y CDS spreads on December 20.
This problem can be solved But how cope with this issue? Apparently, we will fail at obtaining CDS quotes
with simple numerical
approaches, ...
with a maturity of 4¾ and 9¾ years after the first three months of an index
contract’s life. A simple approach is to interpolate the spread curve for CDS
contracts, usually based on quotes for 1-, 3-, 5-, 7- and 10-year contracts.
INTERPOLATING THE CDS CURVE IN ORDER TO OBTAIN SPREADS FOR CONTRACTS WITH TAILOR-MADE TERMS*
120 120
2 3
s = 10.529 + 17.755 t -1.069 t + 0.023 t
100 100
80 80
CDS spread level
CDS spread level
74.8 73.2
60 60
40 40
20 20
0 0
4.75 4.75
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
term term
* We use linear interpolation in the left chart and a third-order polynomial in the right chart. Source: HVB Global Markets Research, Mark-It
... either by using a linear The left chart above depicts how to obtain a spread level approximation for a
interpolation procedure ...
term of 4¾ years, applying a simple linear interpolation procedure. 3Y and 5Y
CDS spreads (denoted by s3 and s5, respectively) act as sampling points for these
purposes. By applying the theorem on intersecting lines, you can derive the
following:
sˆ4.75 − s3 4.75 − 3 4.75 − 3
= ⇔ sˆ4.75 = s3 + ⋅ (s5 − s3 )
s 5 − s3 5−3 5−3
For DCX we obtained an approximation of about 74.8 bp based on CDS spreads


October 5, 2004
observed on August 11, 2004. While implementation of this approach is fairly

straightforward, it utilizes only two of five observed spread levels and thus
partly ignores available data. Nevertheless, the method yields reasonable results
for the illustrated DCX case.
... or by fitting a regression Unlike linear interpolation, fitting a regression equation for the whole curve
equation for the whole curve
takes all available spread levels into consideration. In the right chart above, we
chose a third order polynomial with the maturity t (and t2, t3) as the independent
variable and fitted it by applying the method of least squares. For the DCX case,
we obtained
sˆ = 10.529 + 17.755 ⋅ t − 1.069 ⋅ t 2 + 0.023 ⋅ t 3 .

Insertion of t = 4.75 then yields a spread estimate of about 73.2 bp which
deviates slightly from the linear interpolation result. The major downside of any
parametrical least squares approach is, that observed values usually do not
coincide with the fitted curve (for t = 5 we obtain 75.4, although the observed
spread is 77.9). Taking a look at the right chart reveals this drawback for the
DCX case, as 3Y and 5Y spreads significantly differ from predicted values. Aside
from this, results alter when choosing another functional form.
Performing the calculation For the sake of accuracy, we suggest to perform the above numerical procedure
steps in the right order
(independent from the chosen interpolation approach) for each name within the
respective index before calculating the weighted average of predicted spread
levels according to our pricing formula. Applying one of these methods above
solely to the aggregated index spread curve may lead to a lack of precision due
to the aforementioned effect that lower spread levels are regularly accompanied
by a decreasing spread dispersion. When choosing a polynomial regression
approach, the curvature of individual spread curves differs from name to name.
Hence, interpolating just the aggregation index spread curve would mean to
discard these peculiarities.


October 5, 2004
ISSUE NO. 3 - THE “QUOTATION BIAS”

Recent considerations must be Finally, previously depicted valuation considerations have to be related to
related to pricing conventions
pricing conventions of DJ iTraxx swap contracts. We always discussed the
appropriate level of such a CDS index product given a set of currently
observed single-name CDS spreads. However, the spread level of a DJ iTraxx
swap (some refer to it as the “strike spread”) is fixed a few days prior to
inception and remains unchanged until maturity of the respective series.
Even if one enters into such a contract at a later date and the single-name CDS
spreads have changed significantly in the meantime, the contract will still carry
the announced strike spread as the premium leg. The intrinsic value of this
unreasonable spread level is initially compensated by an up-front payment. This
procedure has been established for the sake of standardization, to which the IIC
is committed. We already mentioned this feature in the basics section.
A quoted spread is only a proxy The main task of this standardized procedure is calculating the amount of the
for calculating the up-front
payment at initiation, ...
up-front payment. Of course, market makers could quote these up-front
payments for a standardized notional amount (e.g. EUR 1 mn). However, credit
investors are used to think in terms of spread levels. This contradiction is solved
by quoting “spreads” that facilitate the calculation of up-front payments based
on a set of crude approximations/assumptions. This could be easily done by
using the CDSW function in Bloomberg. But are these quotes still meaningful if
they are subject to simplifying assumptions? What is the link to our calculated
weighted average incorporating single-name CDS spreads with a possible
adjustment for a maturity mismatch?
CALCULATING UP-FRONT PAYMENTS

... which is not required to The assumptions that are necessary to calculate the upfront payment (the
correspond with our calculated
fair spread
respective fields are highlighted in the left CDSW screenshot below), are:
HOW THE PRESENT VALUE (AND UPFRONT PAYMENT) OF A DJ ITRAXX SWAP IS CALCULATED
Source: Bloomberg
– The JPMorgan valuation model is chosen to calculate the market value.

– The Euro Swap Curve (EU BGN Swap Curve), administered by Bloomberg, is
used as a risk-free term structure of interest rates.


October 5, 2004
– The calculation is based on a flat spread curve assuming quarterly premium

payments, i.e. the quoted “spread” is used for every single maturity.
– The day count convention is fixed to act/360, which is in line with contract
terms of DJ iTraxx swaps.
– A constant recovery rate of 40% is assumed.
Assuming a flat spread curve Stipulating the JPMorgan valuation model for calculation purposes seems to be
and a constant recovery rate of
40% is not in line with single-
arbitrary. However, this should cause no bias owing to the fact that this model is
name CDS pricing market standard for single-name CDS contracts as well. But assuming a flat
spread curve and using a constant recovery rate of 40% is at odds with the usual
procedure to price single-name CDS contracts. This is demonstrated in the
CDSW screenshot on the right above. In reality, credit spread curves are
typically upward sloping as the DCX example shows (see section “Spreads”). In
addition, it is possible to adjust the recovery rate to the market view. At
inception, the spread of an “at-market” CDS (see section “Deal Information”) is
usually fixed to a level that adjusts the initial net present value (“Principal”; see
section “Calculator”) of the contract to zero.
PROBLEMS ARISING FROM THE FLAT CURVE ASSUMPTION

An example depicts the Let’s focus on the problem arising from assuming a flat spread curve,
existence of a quotation bias
keeping the (expected) recovery rate constant at 40%. Suppose a
degenerated index (5Y) consisting of only one name X whose spread is 200
bp (5Y) at inception (alternatively think of an index whose constituents offer
a uniform spread level).
PV-neutrality is the starting Due to the composition of the index, there is no “dispersion bias” as described
point
before. Thus, fixing the index spread to a level of 200 bp requires no upfront
payment for a counterpart (ignoring bid-ask spreads). Consequently, having a
position in a single-name CDS contract is equivalent to a swap contract in the
(degenerated) index. Now suppose a sudden surge in the CDS spread curve of
name X shortly after inception with 5Y level quoting now at 300 bp,
accompanied by a spread curve slope of 50 bp per year (e.g. 10Y spread level
would be 550bp). Applying Bloomberg’s CDSW calculator to the single-name
CDS (governed by a risk-free interest rate curve from August 18, 2004) now
yields a present value of EUR 429,141.10 given a notional amount of 10 mn EUR
(about +4.29%). The index product should have the same present value, as it is
identical to the single-name CDS. However, when applying the 5Y CDS level (300
bp) as quoted “spread” for the index product, we obtain a present value of only
EUR 416,058.72 (about +4.16%). The target value of EUR 429,141.10 is
achieved by quoting a “spread” of 303.27 bp. Thus, we have a quotation bias of
3.27bp within our framework.
There is no quotation bias as To elaborate on the issue of the quotation bias we firstly focus on a setting
long as the strike spread equals
the market spread
where the strike spread equals the market spread. The latter is calculated as a
weighted average of single-name CDS spreads with a possible adjustment to
maturity mismatch. Working with a flat instead of an increasing par spread
curve (as regularly observed) leads to a reduction of the (absolute) present value
of the premium leg. This is due to lower survival probabilities especially for
earlier premium payment dates. However, implementing a flat spread curve
instead of an upward sloping one does not change the present value of the whole
contract, as the strike spread equals the fair market spread. Therefore, no


October 5, 2004
quotation bias occurs.

The relative change of In this special case, the present value of the default leg drops to the same degree
premium and default leg PV’s
is rather stable
as the premium leg owing to a constant value (zero!) of the CDS contract. In
other words, relative present value changes of both CDS legs coincide. This
behavior strictly holds when the strike spread equals the market spread, but it is
somewhat true in other cases as well. We confirmed this theoretical conclusion
with numerical examples. The magnitude of such a relative drop in present
values (2,08% in our numerical example above) is substantially explained by the
slope of the market par spread curve, which is another important insight.
CONSEQUENCES OF DIVERGING STRIKE AND MARKET SPREADS

If strike and market spreads But what are the consequences if strike and market spreads diverge? Let’s
diverge, a uniform relative PV
drop for both legs is
assume a setting where the strike spread lies beneath the current CDS level (as
accompanied by absolute PV in our example above). In this case, the default leg PV exceeds the present value
change of the premium leg. Given the same relative drop of present values when
switching to a flat spread curve, the default leg deteriorates more than the
premium leg. This technical calculation effect can only be compensated by
quoting a higher “spread” level. The reverse is true for a setting where the strike
spread either exceeds the market spread level, or where we have downward
sloping par spread curves (latter is currently rather unrealistic).
We provide a simple method to According to our rationale pointed out above, there is an easy way to calculate
evaluate the quotation bias
an approximation for the quotation bias. A simple simulation procedure with
Bloomberg’s CDSW tool can be utilized to calculate the effect when switching
from the true par spread curve to a flat one.
THE EFFECT OF SWITCHING FROM THE TRUE PAR SPREAD CURVE TO A FLAT ONE
Source: Bloomberg
How big is the market value We stick with our former numerical example to depict the procedure of how to
discrepancy when switching
from the true par spread curve
estimate the quotation bias. Both figures above show the calculation of CDS
to a flat one? market values (present values) either (1) by using the true par spread curve or
(2) by using a flat spread curve subject to the true par spread level for the
particular maturity. Besides, calculations are based on the same contract
specifications of the underlying CDS or index product. Thus, the resulting
market value discrepancy of EUR 13,041.54 (EUR 428,336.28 minus EUR


October 5, 2004
415,294.74) is solely attributable to the application of a flat spread curve.

Secondly, we have to convert As aforementioned, the quoted “spread” level has to be adjusted to bring the
the PV difference to a bp term
present values of both settings in line. In other words, the question is how to
convert the observed PV difference into basis points. The easiest way is to make
use of the calculated value “Sprd DV01” which gives a gauge of market value
sensitivity assuming a setting where the spread curve performs a parallel
upward shift of 1 bp. Just divide the discrepancy in market values by this
sensitivity (rather use the figure from the screenshot on the right). This
generates:
13,041.54 EUR
≈ 3,268 bp
3,990.63 EUR / bp
As you can see, this is a very good approximation for the correct value of the
quotation bias obtained in our former example.
DJ iTraxx swaps currently show Based on these considerations, we estimated the current magnitude of this bias
small quotation biases
for the first series of DJ iTraxx Indices. For simplicity reasons, we assumed
linear par spread curves set up by using quoted 5Y and 10Y spread levels
observed on August 30, 2004. As can be seen, the quotation bias is negligible at
present. However, if we observed a blowout of credit spreads after inception,
this would lead to a reasonable discrepancy between the strike spread and the
average credit spread, leading to a considerable quotation bias. If so, it should
be taken into account.
THE QUOTATION BIAS FOR THE DJ ITRAXX BENCHMARK AND ITS SUBINDICES
5Y contracts 5Y contracts 10Y contracts 10Y contracts

current quotation current quotation
spread dev. bias spread dev. bias
DJ iTraxx Benchmark (125) -7.25 -0.01 -8.50 -0.05
Non-financial corporates (100) -4.00 -0.01 -3.50 -0.03
Auto & auto parts (10) -1.00 0.00 +2.00 +0.02
Consumers (30) +2.00 0.00 +6.00 +0.05
Energy (20) -20.00 -0.03 -23.50 -0.13
Industrials (20) -7.00 -0.01 -7.00 -0.05
TMTs (20) -5.00 -0.01 +1.00 +0.02
Senior Financials (25) -8.00 -0.01 -10.00 -0.04
Subordinated Financials (25) -15.00 -0.03 -13.00 -0.11
DJ iTraxx HiVol (30) -12.50 -0.03 -9.50 -0.08
DJ iTraxx Crossover (30) -17.50 -0.04 -13.00 -0.11


October 5, 2004
SUMMARY
Every DJ iTraxx investor should As illustrated on the last few pages, three different topics have to be taken
be aware of when to consider
a pricing bias
into consideration when trying to calculate a theoretical spread for DJ
iTraxx swap contracts:
– The “dispersion bias” is based upon the fact that a DJ iTraxx swap pays a
uniform spread level for all index constituents, while regular CDS contracts
pay individual and therefore different spread levels. However, this problem
can be resolved by using a weighted instead of an unweighted average of
single-name spread levels, which leads to a significant discrepancy in
calculation, especially when the dispersion of individual spreads is large. As
long as the spread dispersion within an index is limited, this bias is negligible.
– The “maturity mismatch” refers to a lack of appropriate single-name CDS
quotations owing to the fact that the rollover of DJ iTraxx indices occurs only
every 6 months, while regular CDS contracts roll quarterly. This problem
always appears with the beginning of the second quarter in an index
contract’s life and can only be solved by using certain interpolation
approaches.
– The “quotation bias” is attributable to initially fixed spread levels, while
entering into a DJ iTraxx swap is always accompanied by an up-front
payment to compensate for the intrinsic value. In order to convert quoted
“spreads” into this amount (via Bloomberg’s CDSW tool), several partial
unrealistic assumptions are necessary. However, the resulting quotation bias
is negligible as long as the difference between theoretical spread and strike
spread level is limited.
However, theoretical spreads Even when calculating theoretical spreads according to the already mentioned
may deviate from observed
quotes
topics, there still may be discrepancies to where the index market trades. These
deviations are often referred to as “basis to theoretical” or “skew”. Recent
history has shown that this “skew” can be quite substantial, e.g. quoted spreads
for the 10Y DJ iTraxx Crossover swap exceeded theoretical levels by about 20
bp, given that market levels are driven by supply and demand rather than by
sophisticated models. From a theoretical point of view, bid offer spreads and
liquidity considerations allow such deviations to a certain extent.


October 5, 2004
ITRAXX-TRANCHES: VALUATION AND MODELING ASPECTS
INTRODUCTION
Tranched DJ iTraxx products Beside the benchmark index and the sector sub-indices, there are also
are quite similar to unfunded
(synthetic) CDOs
tranched DJ iTraxx products available. These are quite similar to unfunded
CDO transactions, but with an important difference: While most CDOs are
illiquid private transactions, specifically designed to fit the needs of a
portfolio manager and bought by buy-and-hold investors, DJ iTraxx
tranches are standardized, tradable and highly liquid.
Tranche positions can be These tranches have a standardized capital structure based on a liquid
hedged without exorbitant
transaction costs
underlying. They are designed to attract both medium-term-oriented investors
and traders injecting liquidity into the market. Because there is a liquid market
for the underlying – on a single-name and index level – the exposure to tranches
can be hedged without exorbitant transaction costs. Since short-selling for the
tranches as well as for the underlying is possible, investors can efficiently
implement arbitrage and relative-value strategies.
DJ iTraxx CDO tranches belong DJ iTraxx CDO tranches belong to the class of single-tranche CDO (STCDO)
to the class of single-tranche
CDOs
structures. In contrast to fully-tranched CDOs with a complete capital structure,
a STCDO is a specific tranche on a specific pool of assets, which an investor
wants to acquire. The CDO manager compiles the reference pool and sells the
risk of the designated tranche (i.e. buys protection), with specified upper and
lower tranche boundaries (attachment and detachment point) to the investor
(i.e. sells protection). Consequently, the CDO manager has to hedge the risk for
all remaining tranches.
CDO tranches offer leveraged From an investor’s point of view, the DJ iTraxx CDO tranches offer leveraged
exposure to the credit market
exposure to the credit market and thus attractive spread levels without the need
to invest in low single-name credit quality. The impact of defaults on the tranche
value depends strongly on the subordination level of the specific tranche. A
mezzanine tranche for example offers significant spread enhancement compared
to the DJ iTraxx benchmark index, accompanied by a protection against
immediate risk of credit defaults. Even if they are not directly affected, also the
value of higher protected tranches can change significantly in case of the
occurrence of a credit event. This is related to the fact that credit protection of
the tranche deteriorates in case of reduced subordination.
Default correlation is a major The main risk factor – besides the underlying credit spreads – is default
risk factor
correlation among the reference entities. A high default correlation indicates a
higher risk of joint defaults, which may erode higher protected tranches of the
capital structure. The value of the tranches and - accordingly - the spread paid
to investors is highly sensitive to default correlation. Therefore, trading CDO
tranches is often referred to as correlation trading. An unhedged position in a
CDO tranche indicates a market view on individual credit spreads and default
risk in the underlying credits and a view on joint default risk in the pool.
Usually, default correlation is an input parameter for a complex model, which
results in a price. For DJ iTraxx tranches it is the other way around. Due to high
liquidity in the market, prices (i.e. tranche spreads) are driven by supply and
demand. Consequently, the level of default correlation can be extracted from
these prices. Thus, the name of the game is “price discovery” (implied
correlation) rather than “model-based pricing”.


October 5, 2004
Providing customized However, due to the standardization of the underlying CDS portfolio, it is rather
protection is rather difficult
difficult to provide customized protection using the DJ iTraxx CDO tranches.
They may not be suitable as a tool to hedge joint defaults of a specific portfolio.
Main players will be correlation-trading desks and credit hedge funds which
actively trade correlation risk. Additional market participants are bank
proprietary desks, bank loan portfolio managers and insurers who want to get
leveraged exposure to credit risk.
The following section gives an overview of tranche products and sheds some
light on important details such as risk factors, price quotation and sensitivities.
The outline of the section is as follows: First of all, we introduce the concept of
default correlation. Second, we show how tranches can be valued using basic
building blocks. Then follows an analysis of the ongoing discussion concerning
“price discovery” vs. “model competition” and of the price quotation
mechanisms (“base correlation” vs. “implied correlation”). We end this section
with the description of the simple “Homogeneous Large Portfolio Gaussian
Copula” model. The mathematical derivation of the valuation building blocks
and the HLPGC model can be found in the Appendix.
CORRELATION: QUOTATION MECHANISMS AND MARKET STRUCTURE

Default correlation: the name Default correlation is the name of the game for CDO tranches. It is the main
of the game for CDO markets
factor which affects prices and sensitivities of tranches. In the current
section, we shed some light on the economic “nature” of default correlation
and quotation mechanisms. We introduce the concepts of compound and
base correlation which are frequently used in the CDO market. And we
highlight some arguments that explain the current structure of market
prices.
Default “correlation”? Default correlation is frequently cited in conjunction with CDO tranches. But
The term default
“dependency” is more precise
what does this concept actually mean? As a first guess, we could analyze the
correlation between spread changes of two credits in order to measure the
default correlation. However, this is not the best approach and it is not market
standard. The confusion may arise due to the term “correlation”, which is
somewhat misleading. A more precise expression would be “default
dependency”.
A motivating example As motivation for the correlation discussion, we start with a simple example,
which helps us ask the right questions. We assume a model economy with just
two competing companies. Let’s further assume that they have a comparable
market share and economic strength, so that we cannot predict which company
will prevail. Any piece of information, which influences the specific market, will
affect spreads of both companies in the same manner. This results in a high
level of spread correlation. But what happens if one of these companies
defaults? Will the other company default as well? Or will this event result in a
strengthening of the surviving company due to a rising market share? The latter
case indicates that a high level of spread correlation does not necessarily mean a
high level of default dependency.
Two sides of credit risk: In fact, a liquid credit product (corporate bond, CDS or a DJ iTraxx index)
1) default occurence
2) spread changes
incorporates two sources of credit risk (leaving interest rate risk aside):
1) The risk of an occurring default, and
2) the risk that the market perception for the default risk changes (in other


October 5, 2004
words: the risk of spread changes).

Consequently, there are two “sources of correlation”: default occurrence risk
and spread risk. The first gives rise to the concept of default dependency or
default correlation, which is central to the valuation of credit portfolio
derivatives.
Copulas and default Following Li (2000), default dependency can be understood either as a discrete
dependency modeling
default correlation or as a survival time correlation. The first is based on a
single period, e.g. one year. The latter is the more general concept, since we can
easily calculate the discrete correlation once we have the joint distribution of the
individual survival times. The mathematical tool to describe joint distributions is
the copula approach. These are functions, which link individual marginal
distributions to a joint distribution. There are several copulas discussed in
literature, which seem to be applicable to credit risk: Gaussian, Student-t, or the
class of Archimedean copulas (see Li (2000) and Schönbucher (2003) for an
introduction to copula models in credit risk). Note that we (implicitly) use the
Gaussian copula in the “Homogeneous Large Portfolio Gaussian Copula” model
(see Appendix for details).
How to measure correlation? The central problem to default correlation is to parameterize the dependency
structure (i.e. the copula). As we already learned, spread correlation is not the
best choice for measuring default correlation. The most frequently used
approach is to estimate asset return correlation within a Merton-type structural
model (e.g. CreditMetrics). Equity returns are often used as a proxy for these
asset returns. The correlation matrix, which can be obtained from a time series
of stock prices, is plugged into a pricing model, resulting in a CDO quotation.
Correlation is actually driven
by the market and not by a
The result clearly depends on the model and on the parameters. In liquid
model markets, prices are usually not driven by models, but by supply and demand (for
a more detailed discussion regarding this topic please refer to the section
below).
COMPOUND CORRELATION VERSUS BASE CORRELATION

Compound correlation or The prices for CDO tranches based on the DJ iTraxx benchmark index are
base correlation?
quoted as (bid/ask) spreads for a tranche with a given maturity (usually 5Y).
There are quotations available for the following standardized tranches:
equity (0-3%), BBB (3-6%), AAA (6-9%), junior super-senior low (9-12%) and
junior super-senior high (12-22%). The numbers in parenthesis give the
attachment and detachment (a.k.a. exhaustion) points of the tranches. The
quotation for the equity tranche is slightly different, since the investor
receives a fixed running coupon of 500 bp and an up-front payment (see
table below).
In case there is a market price (= tranche spread), we can convert it into an
average pool correlation using the simple HLPGC model, which can be
considered a kind of market standard for this purpose. This is called compound
or implied correlation and most CDO arrangers quote it in tandem with tranche
spreads. Beside the compound correlation, there is another correlation gauge,
which is frequently used in conjunction with CDO tranches: base correlation. We
will elaborate on this concept and other aspects of price quotation using a
numerical example.
An example In the following, we assume 5Y single-tranche CDO contracts with an underlying
index value of 35 bp and a risk-free interest rate of 0%. We generate results by


October 5, 2004
using the HLPGC model (please refer to the Appendix for more detail). In the
table below, we present exemplary data for long positions (selling protection) on
the specified tranches.
SPECIFICATION FOR EXAMPLE CDO TRANCHES, ON A CDS INDEX WITH AN AVERAGE SPREAD LEVEL OF 35 BP
Tranches Borders Spread Compound Base Fair Spread- Delta** Delta** ?V° ?V° ?V°
(in %) correlation correlation Spread leverage* (idx adj) (-5bp) (+5bp) (+10bp)
Equity 0-3% 20.40% 26.97% 26.97% 1,027 29.34 -88.6 -17.8 4.7% -4.3% -8.3%
+500 bp°° bp
BBB 3-6% 106 bp 9.19% 37.63% 106 bp 3.03 -42.9 -8.6 1.9% -2.3% -4.9%
AAA 6-9% 43 bp 17.12% 45.27% 43 bp 1.23 -15.2 -3.1 0.7% -0.8% -1.8%
Junior 9-12% 27 bp 23.72% 51.15% 27 bp 0.78 -9.0 -1.8 0.4% -0.5% -1.0%
Senior 12-22% 17 bp 34.90% 64.35% 17 bp 0.50 -4.8 -1.0 0.2% -0.3% -0.5%
• The spread leverage is calculated as fair-spread / index-spread (35 bp).
** Delta is calculated as the change of tranche value with respect to a change in the underlying index using given levels.
The index adjusted delta is calculated as delta(tranche) / delta(index).
° ∆V (-5 bp) gives the change in tranche value, if the underlying index changes by -5 bp.
°° According to market standards, the price for the equity tranche is quoted as an up-front payment of 20.4% and a running spread of 500 bp.
In the first two columns, tranches are named and specified by their lower
(attachment point) and upper (detachment point) borders. Then we state the
tranche spread (note that the equity tranche is quoted as a running spread of
500 bp and as an up-front payment of 20.4%), followed by compound and base
correlation. The rest of the table gives further properties of the tranches, such as
delta sensitivities.
Compound correlation is The compound correlation is calculated as the (implied) correlation, which re-
calculated as the (implied)
correlation
prices the specific tranche in terms of the HLPGC model. For example, the
spread of 106 bp for the BBB (mezzanine) tranche can be reproduced using an
average pool correlation of 9.19%. The base correlation is the (implied)
correlation of the corresponding equity tranche with the same detachment point
(upper border). In this case, it is the 0-6% tranche, which has an implied
correlation of 37.63%. Please note that compound and base correlation for the
equity (0-3%) tranche are obviously identical. (A short hint to our notation: we
use the terms “compound correlation” and “base correlation” as stated above,
whereas “implied correlation” is used in a more general sense, since both
correlation figures are implied in prices.)
COMPOUND CORRELATION … … AND BASE CORRELATION

70.00% 70.00%
60.00% 60.00%
50.00% 50.00%
40.00% 40.00%
30.00% 30.00%
20.00% 20.00%
10.00% 10.00%
0.00% 0.00%
0%-3% 3%-6% 6%-9% 9%-12% 12%-22% 0% 5% 10% 15% 20% 25%


October 5, 2004
The characteristic structure of In the charts above, we highlight compound and base correlation for the
the compound correlation is
frequently called the
indicated example. The characteristic structure of the compound correlation is
correlation smile frequently called the correlation smile. To plot it, we use a column diagram and
not an interpolated scatter chart, since each tranche has to be viewed
separately. We can compare the correlation levels among the tranches, and can
conclude for example that the compound correlation for the equity tranche is
higher than the one for the BBB tranche, but there is no meaningful way of
interpolating the correlation of the 0-3% and the 3-6% tranche. Since both –
upper and lower – points are shifted, it is unclear, what happens to correlation
between these two points.
Base correlation figures can be The situation is different for base correlation. Since only the detachment point is
interpolated
shifted and the attachment point for all equity tranches is kept fixed (at 0%),
there is a meaningful way of interpolating between two points. The 4.5% base
correlation is between the 3% and 6% base correlation. Thus, once we know the
base correlation for a few tranches, we can calculate the correlation for all other
equity tranches between the lowest and the highest point by interpolation.
Expected loss connects Compound correlation and base correlation are related to each other via the
compound and base
correlation
expected tranche loss. The expected loss of a compound tranche is the difference
of expected losses of the corresponding equity tranches, adjusted by the
thickness of the tranches (Please refer to the Appendix “Payoff function for base
correlation” for mathematical details). This relation can be used to calculate the
base correlation from the sequence of compound correlations in a bootstrapping
algorithm: for the first tranche, compound and base correlation are the same.
For the second tranche, calculate the market implied expected loss of the “base
tranche” as the (thickness adjusted) sum of expected losses of the first two
tranches and solve for the correlation parameter, which reproduces this
expected loss. Thus, we receive the base correlation and the expected loss of the
second equity tranche. To analyze the senior tranches, we repeat this procedure
with the next compound tranche.
Base correlation allows to price Using the equation in the appendix, we can also infer the price of an off-the-run
off-the-run tranches …
tranche, such as a 4-8% tranche. This is clearly an advantage of this quotation
mechanism. From compound correlation we cannot derive this price directly.
… but compound correlation But compound correlation is not useless. In fact, the curious structure says a lot
reveals the “smile”
about the market’s view on risk, which we cannot derive directly from base
correlation. Let us briefly discuss the smile structure (for more details please
refer to the section below on sensitivities of CDO tranches). A high level of
correlation shifts default probability from the equity to higher tranches in the
capital structure. Thus, a high level of correlation results in a reduced equity
tranche spread. The dependency of the mezzanine tranche on the level of
correlation is less obvious. For a low correlation, the tranche spread increases
with increasing correlation, but for higher levels of correlation it decreases with
increasing correlation. The spread dependency of senior tranches is contrary to
those of equity tranches. A high level of risk means a high level of clustered
(large) losses, which might also affect senior tranches. Thus, a high correlation
means higher spreads for senior tranches.


October 5, 2004
CORRELATION: WHY IS THERE A SMILE?

Correlation smile: On the back of this analysis, the shape of correlation structure is
exploiting investors appetite
for risk in equity and
understandable. Since there is a high demand in the market to sell
mezzanine tranches? protection in equity and mezzanine tranches, equity correlation goes up
(which reduces the tranche spread) and mezzanine correlation goes down
(which reduces the corresponding tranche spread as well). There is little
demand for higher protected tranches. Thus, correlation is high for this
side. Due to the supply-and-demand imbalance across the capital structure
(high demand for low tranches, low demand for high tranches), equity and
mezzanine tranches are (relatively) more expensive from an investor’s point
of view than other tranches.
There is another argument why the right branch of the correlation chart is
upward sloping. Since senior tranches are very well protected, the risk of a
default event which might hit them is negligible. The fair spread would be
therefore approximately zero. But since there will not be anybody out there who
would like to take the risk of joint defaults for zero reward, the correlation has
to be at a level where these tranches pay at least a basis point or so.
How to model the smile Concerning the modeling aspect, there is an ongoing discussion in the financial
community how to model this correlation “smile” in compound correlation
within a more sophisticated model. Hull and White claim that a single-factor
model with two student-t copulas (with 4 degrees of freedom each) - one for the
market factor and one for the individual default probabilities - is able to
reproduce this shape. (We will address this topic and the issue of the implied
default distribution in the next release of the DJ iTraxx special publication).
At-the-money correlation: Finally, there is another interesting topic which supports the notion of base
compare correlation across
time
correlation: the comparison of correlation levels across markets or over time.
This can be difficult since the average spread in the underlying pool can change
over time. So it is not clear, whether it makes sense to compare the compound
correlation of a mezzanine tranche for an index spread level of 30 bp with the
corresponding one in a situation in which it is 60 bp. JP Morgan claims that it
can be compared using the so-called concept of at-the-money (ATM) correlation
in the context of base correlation. This essentially means that the base
correlation is economically viewed as a (stationary) function of the expected loss
and not as a function of the detachment point. Just imagine that we convert the
x-axis in the right chart above from the detachment point into expected loss. We
can do this for different points in time (or different countries) and could compare
the two ATM correlations. This allows us to decide whether correlation is cheap
or expensive. However, in our opinion, the time series of tranche prices is yet
too limited for a reliable conclusion about the concept of ATM correlation.
Correlation dynamics and The question of “How does implied correlation change if the average spread
hedging sensitivities
level of the underlying pool changes?” is also relevant in terms of delta-hedging.
In order to calculate the sensitivity of a tranche with respect to the index spread,
one shifts the index spread and keeps everything else fixed. But what does
“keeping everything else fixed” actually mean, if the correlation parameter is a
function of the underlying spread level? Does one has to keep compound, base
or ATM-correlation fixed? This is one of the exciting topics, which affects CDO
arrangers, since they have to hedge the exposure. But there is no satisfying
conclusion yet. However, for investors seeking leveraged spread exposure and
who do not want to trade correlation actively, this is a side issue.


October 5, 2004
Implied correlation is Last but not least, the implied correlation, whether it is compound or base
influenced by various factors
correlation, is a complicated function of several economic quantities, since it
accounts for “everything else” which is not captured by the HLPGC model, but
which is important to the market. To name just a few factors, it will be
influenced by the individual pairwise correlation, but also by the spread
dispersion in the pool, or by different recovery assumptions.
HEDGING CORRELATION PRODUCTS: SENSITIVITY QUOTATION

Hedging needs from CDO Following, we briefly introduce the most important tranche sensitivities,
managers drive the underlying
index market
which are usually quoted on market maker screens. The striking point is
that the leverage of tranche products with respect to the underlying can be
really big. Hedging needs from CDO managers will therefore drive the
underlying index market.
For a more detailed analysis of tranche sensitivities please refer to the
section “DJ iTraxx tranches”.
Hedging needs for iTraxx The negative delta values in the table above indicate that the long credit risk
equity tranche investors is ~18
times the underlying index
positions of the tranches lose value in a spread-widening scenario. The equity
tranche features the highest sensitivity and offers the highest spread-leverage. A
spread widening in the underlying index of 1 bp results in a loss in tranche
value of 88.6 bp. Under the given scenario, the corresponding sensitivity (PV-
change) of the underlying index is approximately –5.0. Thus, to hedge the
spread risk of the equity tranche with a position in the underlying index we need
17.7 (= 88.6 / 5.0) units of the index.
A EUR 10 bn BBB-tranche Thus, an investment of EUR 100 mn in the equity tranche triggers hedging
investment creates hedging
needs totaling ~ 25% of the
needs of EUR 1.77 bn (!) in the underlying index. For the BBB-tranche, the
outstanding amount of the corresponding amounts are still 8.6 times the credit exposure. Given that the
iBoxx NFI iBoxx non-financials universe has about EUR 350 bn bonds outstanding (EUR
600 bn for the iBoxx corporates, incl. financials), we would need about 25% of
the iBoxx NFI volume to hedge EUR 10 bn in BBB tranches. Consequently,
buying protection for a EUR 10 bn BBB-tranche investment would have a
significant spread impact.
Break-even spreads Given these huge sensitivities, break-even spread changes are quite low. We can
approximate the break-even spread via the delta-sensitivity and the fair-spread.
As an example, the 3M breakeven for the equity tranche is about 3 bp (10.27% *
0.25Y / 88.6). To generate zero return, an equity investor can suffer from a
spread widening of 3 bp in the underlying index if he earns a tranche spread of
10.27% quarterly.
A 3 bp cushion seems not too comfortable given the current spread volatility in
the DJ iTraxx (the DJ iTraxx Europe spread tightened by about 5 bp during the
second half of August 2004).
P&L impact of leveraged From the table above (page 27), we can also extract the P&L impact in case the
exposure significantly exceeds
that of plain-vanilla credit
underlying index spread changes by –5 bp, +5 bp and +10 bp, respectively.
investments Given an index level of the DJ iTraxx benchmark of 35 bp, an unhedged position
in the equity tranche will generate 4.7% return if the index tightens 5 bp to 30
bp. On the other hand, the position will lose 4.3%, if the index widens by 5 bp to
40 bp. That said, the P&L amplitude of leveraged credit exposure significantly
exceeds that of plain-vanilla credit investments. (Note that we calculated all
sensitivities by keeping the compound correlation fixed. Please refer to the
section above for an assessment of the accuracy of this approximation).


October 5, 2004
VALUATION WITH BUILDING BLOCKS

The expected survival amount The basic building block for valuation of credit risky products is the (time-
is the basic building block for
valuation of credit risky
dependent) expected survival amount. It gives the fraction of a cash flow
products which is expected to survive, and is related to the expected loss simply by S
= (1-EL) = (1-PD*LGD). Both quantities involve the default probability and
the loss-given-default.
A simple example, … We start with a simple example: we assume an exposure of EUR 1 mn to a
counterpart with a default probability of 1% p.a. In case of default, we expect a
recovery rate of 40%. The credit has a maturity of 1 year. The expected loss can
be calculated by 1% * 40% = 0.4%, or 4.000 EUR. The corresponding surviving
amount is (1-0.4%) = 99.6% or 0.996 mn EUR. Since the probabilities involved
in this calculation depend on the time horizon, the expected survival amount is
time dependent as well. Moreover, for valuation purposes, we have to use risk-
neutral probability, rather than physical (or real-world) probabilities.
… which results can be applied So far, the results of our analysis can be applied to simple credits as well as to
to simple credits as well as to
complicated credit derivatives
complicated credit derivatives, such as portfolio tranches. For all of them we
have to calculate the time-dependent expected loss or expected survival amount.
Most of the complexity is hidden in these quantities. The difference between the
credit and a CDO tranche is that the first references only to the credit risk of a
single counterpart. The latter is defined with respect to a pool of assets and is
conditional on the loss boundaries of the tranche, e.g. an equity tranche might
cover 0%-3% of all losses.
Valuation of the premium leg In addition, we know two basic structures in CDO as well as in single-name
transactions: funded (bond-like) and unfunded (swap-like) structures. The latter
involves the calculation of a premium and a default leg, just as for plain-vanilla
CDS. The valuation of the premium leg is similar to the funded structure, while
the valuation of the default leg is a little more complicated.
Value of a funded investment The value of a funded investment in a specific tranche (with specified upper and
in a specific tranche
lower bounds) can be calculated via the following mathematical expression,
which is basically discounting all (expected) cash flows with the risk-free zero-
coupon bond prices (the individual terms will be explained below).
n
VBond = ∑c Bond ⋅ δ i ⋅ S (t , Ti ) ⋅ B(t , Ti ) + S (t , Tn ) ⋅ B(t , Tn )
i =1
Separately pricing of the The valuation of an unfunded investment (selling protection, long credit risk) in
premium and default leg
necessary
a specific tranche (with specified upper and lower bounds) is performed by
separately pricing the premium and the default leg. For the default leg, we
assume that repayments of defaults only occur at coupon dates.
VSwap = V PL − V DL
n
V PL = ∑c Swap ⋅ δ i ⋅ S (t , Ti ) ⋅ B(t , Ti )
i =1
n
V DL = ∑ [S (t, T ) − S (t, T )]⋅ B(t, T )
i =1
i −1 i i
With δ i = Ti − Ti −1 the accrual period between two coupon dates, c the coupon of
the tranche (with cSwap the coupon in an unfunded and c Bond the coupon in a


October 5, 2004
funded structure) and B(t , Ti ) the (risk free) zero-coupon bond price for maturity
Ti . The expected survival amount S (t , Ti ) up to time Ti is related to the expected
loss via the simple relation S (t , Ti ) = 1 − EL(t , Ti ) . In the following we focus on the
expected loss EL(t , Ti ) up to time Ti .
THE EXPECTED LOSS

The expected loss … The value of the premium leg is given by discounting the expected premium
cash flows (equals coupon size times expected survival amount) with the
risk-free zero-coupon bond prices. This is similar to the bond-like structure,
except of the missing repayment of the notional amount (cf. that the term
S (t , Tn ) ⋅ B (t , Tn ) is missing in the formula for the premium leg).
The value of the default leg is given by discounting the expected payments the
protection seller has to pay in case of defaults. We calculate them as the change
in the expected survival amount S (t , Ti−1 ) − S (t , Ti ) . Note that here we assumed
that default payments are settled at the fixed coupon payments dates. S (t , Ti ) is a
decreasing function of T , therefore the difference S (t , Ti−1 ) − S (t , Ti ) is also
positive. Hence the swap value is given as the difference between premium and
default leg.
… has to be within the upper The derivation of the mathematical expression for the expected loss is given in
and lower tranche boundary,
…
the Appendix. Here, we think it is illustrative to provide an insight on the
parameters, which influence the value of the expected loss. Obviously, this value
will depend on the upper and lower tranche boundary a and b . The expected
loss has to be within these boundaries. Any loss below a does not affect the
tranche capital at all. And all losses exceeding the upper boundary b do not
have an additional impact on the tranche, since all invested capital had been
eroded.
r
… and depends on spread Additionally, the expected loss depends on rthe credit spreads s of the
levels, recovery rates and
underlying entities, on the recovery rates R and finally on the default correlation
default correlation r
parameters ρ . The credit spread and the recovery rates are inter-related, since
the spreads involve both the probability of default and the loss-given-default,
which is 1 − R . Thus, we need a two-stage algorithm: 1) extract the default
probability from credit spreads (using R ) and 2) calculate the payoff in case of a
default (using R ).
The vector notation used above indicates that there is more than one parameter.
Since the expected loss will, as indicated above, also depend on the time horizon
r r r
( )
we can use ELba t , Ti , ρ , s , R to describe all dependencies.
We consider default In general, there are m spreads and m recovery rates for a pool of m credits,
correlation rather than spread
one per credit. Furthermore, there should be m ⋅ (m − 1) bilateral default
correlation r
correlation parameters, i.e. the vector ρ should be a correlation matrix. As
indicated above, we consider default correlation rather than spread correlation
in this context. Spread correlation is usually neglected in CDO pricing.
And to make things even more complicate, the spreads (and of course the other
parameters as well) could be time-dependent. For spreads, for example, it is
quite common to use term-structures.
The mathematical structure of Having said this, we are still not able to price a contract, since we do not know
r r r
the expected loss function
( )
the mathematical structure of the expected loss function ELba t , Ti , ρ , s , R and we
do not know the appropriate parameters. Perhaps you will be worried from
where one could get all these parameters (in fact it will be


October 5, 2004
m(m − 1) + m + m = m(m + 1) parameters from bilateral correlation, individual

spreads and recovery rates). It is unrealistic to estimate this amount of
parameters. We will have to think hard about a model which is easy to
understand, stable and parsimonious regarding the number of parameters.
In the appendix we show the mathematical derivation of a simple model. For the
moment, let us assume that we have an appropriate function and know all
necessary parameters.
The value of a funded CDO tranche written in terms of the expected loss function
is given by
r r
∑ δ ⋅ [1 − EL (t, T , ρ , s , R )]⋅ B(t, T ) + [1 − EL (t, T , ρ , s , R )]⋅ B(t, T )
n
r r r r
Vab Bond = c Bond ⋅ i
b
a i i
b
a n n
i =1
The value of an unfunded CDO tranche (selling protection, long credit risk
protection) written in terms of the expected loss function is given by
r
∑ δ ⋅ [1 − EL (t , T , ρ , s , R )]⋅ B(t , T )
n
r r
Vab Swap = cSwap ⋅ i
b
a i i
i =1
r r r r r r
∑ [EL (t , T , ρ , s , R ) − EL (t, T )]
n
− b
a i
b
a i −1 , ρ , s , R ⋅ B(t , Ti )
i =1
We obtained these equations from the previous ones simply by substituting the
r r r
( )
S (t , Ti ) terms with ELba t , Ti , ρ , s , R . Using the definition of the value of a risky
annuity Aab (risk duration)
r
∑ δ ⋅ [1 − EL (t , T , ρ , s , R )]⋅ B(t , T )
n
r r
Aab = i
b
a i i
i =1
we can calculate fair spreads of the funded transaction ( Vab Bond = 1 ) and of the
unfunded transaction ( Vab Swap = 0 )
r r r
c Bond =
[ (
1 − 1 − ELba t , Tn , ρ , s , R B(t , Tn ) )]
Aab
r r r r r r
∑[ ( ) ( )]
n
ELba t , Ti , ρ , s , R − ELba t , Ti−1 , ρ , s , R ⋅ B(t , Ti )
cSwap = i =1
Aab


October 5, 2004
SIMILARITIES TO ANALYZING SINGLE-NAME CREDIT RISKS

The result of this analysis is The result of this analysis is similar to the valuation approach for risk-free
similar to the valuation
approach for risk-free bonds …
bonds: discount all cash flows with the prices of zero coupon bonds of the
same maturity. The bond price, which is used for discounting, is adjusted to
r r r
( )
reflect the credit risk. The adjustment factor is given by 1 − ELba t , Ti , ρ , s , R .
This is exactly the same approach as for normal single-name credit risks.
… while there is a striking There is a striking difference in the interpretation of the adjustment factor
difference in the interpretation
of the adjustment factor
between a single-name risk and a portfolio risk. The expected loss for a single
name is just a descriptive quantity, it cannot be observed on a single-name level
in reality. The (model) economy has just two states: the company either defaults
(payoff = R<1) or it survives (payoff = 1). The expected loss is a probability
weighted average of the payoff. But in reality an average value across these two
states in not observable. For a portfolio of risky assets this is different. The
model economy has a lot more possible states. In this case (through
diversification) the expected value of the payoff cash flows need not, but can be
observed in reality.
The central quantity for CDO At this stage, it is clear that the central quantity for CDO valuation is the
r r r
valuation is the expected loss
function
( )
expected loss function ELba t , Ti , ρ , s , R . Once we have a model and all necessary
parameters, we can calculate the value and the fair spread of a tranche. This is
especially the case in model-driven (illiquid) markets. In case of liquid markets,
the value (and thus the fair spread) of CDO tranches is driven by supply-and-
demand. In this case, the model can be inverted: plug in the fair spread, derive
the expected loss from it and back out the parameters from it. This leads us to
the concept of implied correlation (see the section “Price discovery vs. model
competition”).


October 5, 2004
PRICE DISCOVERY VS. MODEL COMPETITION

A proper model results in One of the interesting topics, which is currently discussed by the financial
sound hedging parameters and
in “good” prices.
community, is “price discovery” vs. “model competition”. This discussion is
closely linked to the maturity of the whole market. In early stages of the
development of a market, prices are derived by models. A proper model,
which covers all (or most of the important) risks results in sound hedging
parameters and in “good” prices.
Standard models would boost Thus, market participants who have access to such a model have a good chance
turnover
to outperform their competitors. Such models are usually complex and only
specialized market participant can handle them, which makes the whole thing
rather intransparent. This obviously puts a limitation on the further
development of the market, since investors will only enter the market if the
value of their position does not depend on obscure models.
Prices are driven by supply and On the other hand, if the market becomes more liquid, prices are no longer
demand rather than by models
driven by the models, but by supply and demand. This means that the real price
is one on which the market participant can agree upon. This adds transparency,
since one can mark-to-market a position without a complicated model and
without model risk. So, does this mean that market rules and models are not
needed? A clear “No!” is the answer. In fact, there is no conflict between price
discovery and model competition. In the current market environment we need
both.
But, we need a model to On one hand, the pricing mechanism for the products has to be flexible enough
identify all risks, …
that prices can fluctuate according to the market consensus. On the other hand,
we need a model to identify all risks and quantify their impact on the prices. In
other words, risk managers have to hedge all risks involved in the transaction,
which they do not want to take. And therefore, they need a model, which is
capable of calculating sensitivities. This is especially true for the arranger of
such single-tranche CDO (STCDO) transactions. Since the arranger just sold the
risk of a specific tranche to an investor, he/she has to hedge the risks of the
remaining capital structure, and therefore need sophisticated models. So model
competition takes place on the side of the arrangers of CDO. The company with
the best model can offer most competitive prices for single-tranche CDO and
thus gets a larger market share.
… and to identify price The investors point-of-view on models is somewhat different. Models should be
inconsistencies and arbitrage
opportunities
simple and robust. The major task is to isolate the relevant risk factor: the
correlation risk. And we need to identify its economic drivers. That will allow us
to compare default correlation across different products, sectors and regions.
And it will help us to find an answer to the question whether correlation is a
new asset class (such as volatility). In addition, we can identify price
inconsistencies and arbitrage opportunities.


October 5, 2004
HOMOGENEOUS LARGE PORTFOLIO GAUSSIAN COPULA MODEL

There are several risk factors The risk factors affecting the value of CDO tranches are default risk (single-
related to CDO tranches
name events and correlated defaults), uncertainty about future default
correlation, uncertainty about the real recovery rate in case of defaults,
uncertainty about the future spreads and spread correlation, and interest
rate uncertainty. The tranche values have different sensitivities on these
risk factors. For example default correlation is the major driver for the
price, while spread correlation has very limited impact.
Most important are … The most important risk factors a model has to incorporate are:
… default risk, … – Default risk: There is an immediate loss for the affected tranche. The
protection seller has to pay the loss and the notional amount of the tranche on
which the coupon is paid is reduced. Higher protected tranches also suffer
due to the reduced subordination.
… spread risk, … – Spread risk: The value of the tranches changes with variation in the value of
the underlying CDS contracts.
… and correlation risk – Correlation risk: The value of the tranches changes with (implied) default
correlation.
Numerous models deal with Numerous models dealing with credit portfolio derivatives (such as CDOs) are
credit portfolio derivatives
discussed in literature. Their focus is on modeling the loss distribution function,
which is the basis for the derivation of the expected tranche loss. The most
general models, which take the individual spreads of all underlying and all
pairwise correlation parameters are too complex to be solved analytically. Thus,
one needs Monte Carlo techniques to get a solution. This is clearly a limitation,
since Monte-Carlo simulation tend to be time-consuming and can be numerically
unstable.
Factor-models significantly The so-called factor-models, such as Gregory/Laurent (2003), Hull/White (2004)
reduce the number of
variables, which drive the
and Andersen et al (2003) significantly reduce the number of variables, which
default correlation drive the default correlation. They are all quite similar: they take individual
spreads as input, default dependency is driven by asset correlation (Merton-type
model), can be formulated as single- and multi-factor models and they can
incorporate all kinds of copulas to model default dependency. And there is
another common property: one still needs numerical procedures to solve for the
loss distribution, since it cannot be solved analytically. The differences are how
these numerical procedures are implemented. The suggestions range from
Fourier-transformation to iterative schemes.
We will not go into detail for these models, but focus on a simple, easy
implementable and stable model: the “Homogeneous Large Portfolio Gaussian
Copula” model (HLPGC). This model can be used to study all important topics,
and we do not need a complicated numerical scheme to solve for the loss
distribution, since there is an analytical expression for it.
The HLPGC model … The HLPGC model dates back to the simple Vasicek model for portfolio credit
risk (Vasicek 1987, 1991). It aims at modeling default dependency and is based
on a simplified firm-value approach (Merton-type model). It has only a few
parameters and can easily be implemented in a spreadsheet. Due to its
simplicity, it is a good candidate for tutorial purposes and serves as a vehicle to
explore basic features of default correlation. On the other hand, it appears too
simplistic to serve as a real pricing engine. However, major market participants
agreed to use it for price quotation. We address this topic in the next section.


October 5, 2004
… refers to a … We start with the most immediate question: What does “homogeneous large
portfolio gaussian copula” mean? The answer to this question already uncovers
a lot about the characteristics and the (crude) approximations of the model.
… homogenous, .. – homogeneous: This addresses the constituting credits of the underlying pool.
It is assumed that they all have the same size, the same time-dependent
default probability, the same recovery rate and the same correlation
parameter. Moreover, the default probability is usually calculated using a
constant “clean spread”. Clean spread means the part of the spread, which
accounts for the default probability. It can be calculated via:
dirty spread = clean spread ⋅ (1 − R ) .
The reason for this assumption is to reduce the number of parameters we
need for the model. And the reduction is significant, especially if the pool is
large (see also next point): instead of m(m + 1) parameters in the general case
we just need three. This means that the number of parameters is independent
of the size of the pool, which is of advantage especially for large pools. The
approach is to take some kind of portfolio average for each parameter. The
economic impact of this approximation is severe. To lift it means to
implement a model, which is significantly more complicated.
… large, … – large: This addresses the size and the granularity of the pool. We assume that
the pool is sufficiently large, i.e. it has infinite constituents. Thus, expected
loss function is continuous. Consider a small pool with 10 equally sized
constituents. Here, the expected loss function is not continuous, since a single
default causes a loss of 10% (we assume 0% recovery).
The advantage behind this approximation is that we can obtain the
continuous distribution functions analytically (i.e. the gaussian distribution,
see next point). Since CDOs are usually written on quite large pools (for
smaller ones nth-to-default baskets are more common) this assumption is too
critical. Lifting this assumption means using a discrete distribution function
(such as the binomial distribution) instead of a continuous function.
– gaussian copula: This topic is the heart of modeling the default dependency
… pool of credits, with default
correlation is being captured structure. With “gaussian copula” we indicate that the default correlation is
by a gaussian copula incorporated via correlated stochastic processes (for quantitative experts:
these are Wiener processes), which involve the Gaussian distribution function.
Here the already mentioned firm-value (Merton-type) approach comes into
play. We’ll come back to this in more detail later on.
This type of default correlation is widely used (it is, for example, similar to the
CreditMetrics approach) and well accepted. One reason is that the correlation
parameters can be derived from stock prices. However, there are numerous
suggestions for other “copulas” to model the dependency structure in default
risk (just google for “copula” and “credit risk” to get an indication). A large
part of academic and industry research was dedicated to this topic.
Criticism of the HLPGC Having said a lot about the characteristics of the model, we try to summarize
some model criticism concerning economic properties, which we neglected, and
whether we can adjust the model to account for them.
– spread dispersion: it is a harsh approximation to assume that all credits in the
pool have the same credit spread and hence the same default probability. The
spread dispersion in the pool will affect the calculated prices (see the section
below about price quotation). Unfortunately, this is one of the basic
assumptions of the model.


October 5, 2004
– bilateral default correlation: the “average pool correlation” approach, which

we adopt by using the discussed model does not account for bilateral
correlation effects. To shed some light on the limitations, we just have to think
about economic scenarios, which involve individual correlations, for example
sector or regional correlation, or company-subcontractor relations.
– term-structure of credit spreads: usually one uses a flat spread level in the
model, but this approximation is not necessary. We can easily calculate the
default probability using an average (!) term structure, if we have one.
– uncertainty of recovery: assuming statistical independence between
recoveries and default probabilities, we can incorporate the recovery
uncertainty in the model without major problems.
– spread correlation: since spreads are modeled as a non-stochastic, average
value, we integrate a spread correlation. However, spread correlation does
not affect prices of tranches in a critical way.
All these topics are daily business of credit analysts and risk managers, which
try to rate individual companies or try to manage concentration risk. This model
assumes economic conditions, under which all this is obsolete. But having said
this, it is striking that this model performs quite well, and it is already a kind of
a market standard for price quotation (see below).
REFERENCES
David Li, “On Default Correlation: A Copula Function Approach”, RiskMetrics working paper, April
2000
Philipp Schönbucher, “Credit Derivatives Pricing Models”, Wiley 2003
Oldrich Vasicek, “Probability of Loss on Loan Portfolio”, Working Paper, KMV Corporation, 1987
Oldrich Vasicek, “Limiting Loan Loss Probability Distribution”, Working Paper, KMV Corporation,
1991
John Hull, Alan White, “Valuation of a CDO and an n-th to Default CDS Without Monte Carlo
Simulation”, Working Paper, September 2004
Jon Gregory, Jean-Paul Laurent “I will survive”, RISK, June 2003
Leif Andersen, Jakob Sidenius, Susante Basu, “All your hedges in one basket” RISK, November 2003


October 5, 2004
SENSITIVITIES OF DJ ITRAXX INDEX TRANCHES

Sensitivity gauges for tranched As seen before, pricing of credit index tranches requires quite more
products are indispensable,
even for investors of open
sophisticated valuation techniques than index swaps. This increased
positions complexity in pricing is generally accompanied by higher difficulties when
trying to hedge index tranche positions. Sensitivity gauges towards relevant
risk factors are a convenient way to accomplish this task.
Needless to say there are many investors in index tranches that just want to take
an open position in order to express their market view (or just to receive the
carry in case of a long position in credit risk) without following a hedge strategy
(a perfect hedge would mean no risk, and no risk would mean no return, except
for arbitrage opportunities). Nevertheless, even an investor in unhedged
tranches should have a notion how his position reacts when risk factors vary.
This is why we now focus on the topic of index tranche sensitivities.
The set of relevant risk factors What are relevant risk factors for tranched index products? First of all, we have
depends on which model you
take
to differentiate between risks that emerge in reality and those incorporated in
the used valuation model. Possibly, there are risk factors, which, for simplicity
reasons, are not covered by the model. Even if the model includes the respective
risk factor, there is no guarantee that reality and model coincide. All this is a
consequence of what is called “model risk”.
Restrictive assumptions in the We already depicted the “homogeneous large portfolio gaussian copula”
HLPGC model leave only a
handful of risk factors
(HLPGC) model as a simple way to model portfolio credit risk. A couple of
restrictive assumptions of this model reduce the number of parameters to a
minimum, leaving only a handful of risk factors:
– A change in the uniform spread level for the index constituents (reflecting the
default probability given the uniform recovery rate)
– A change in the uniform correlation parameter
– A change in the uniform recovery rate
Apparently, there are many other risk factors in reality, particularly those which
are due to the dispersion of single-name spreads, recovery rates, and pairwise
correlations, respectively. For example, it is definitely useful to know how the
present value of an index tranche reacts, if we assume an individual credit
spread blowout in contrast to a pool-wide spread movement. Sensitivities
concerning different risk factors can only be calculated on the background of a
pricing model. Thus, this kind of sensitivity cannot be analyzed within the
HLPGC model. Nevertheless, we think that presenting sensitivities based on the
HLPGC model gives a good indication of how DJ iTraxx Index tranches work in
practice.
Default risk is linked to spread A risk factor which has not been mentioned yet is the default risk. A default
risk, but hedging both of them
is normally a tough task
causes an immediate loss for the equity tranche and reduces the subordination
of other tranches, thus lowering the present value of such a contract. Usually,
this kind of risk is presented independently from spread risk. It should be
emphasized that spread risk is nothing else but a latent expression of default
risk. Both types of risk are inextricably linked together given a fixed recovery
rate. Nevertheless, asking both questions of how an index tranche behaves when
(1) spreads widen and (2) defaults occur is legitimate. Measuring these sensi-
tivities very often discloses that they are not complementary. An effective hedge
against default risk involves an open position in spread risk and vice versa!


October 5, 2004
The passage of time is no risk, Option pricing theory often refers to what is called a “theta risk”, which
but has to be taken into
account when establishing a
measures the change in present value of a contract solely due to elapsing time
hedge position until maturity. We like to point out that time is no risk factor, as time will pass
anyway. But the theta measure is good for credit risk portfolios, because it
particularly incorporates the carry issue to our examination. In the majority of
cases, a hedging strategy for a tranched product featuring a negative carry
should mostly be avoided.
SENSITIVITIES WITH RESPECT TO A CONTRACT’S PRESENT VALUE AND FAIR

SPREAD LEVEL
Calculation of sensitivities is As aforementioned, our calculation of sensitivities is based on the HLPGC
based on the HLPGC model
with a unique starting point
model, which will give us a reasonable insight of how DJ iTraxx tranches
for all analyses work. For the sake of comparability, we always start with the same
parameter setting.
We focus on DJ iTraxx Europe tranches with a maturity of 5 years. Let’s
consider a flat term structure of interest rates with an interest rate level of zero.
Moreover, a uniform recovery rate of 40% (R = 0.4) associated with a spread
level of 60 bp (“clean” spread level is then 100 bp). If we stipulate a uniform
correlation parameter Rho of 0.3 as a starting point, the tranches should have
the following fair spread levels:
FAIR SPREAD LEVELS OF DJ ITRAXX EUROPE TRANCHES (5Y)
Tranche Equity BBB AAA Junior Senior

Losses 0 – 3% 3 – 6% 6 – 9% 9 – 12% 12 – 22%
Fair Spread 1194 bp 467 bp 219 bp 114 bp 37 bp
We treat the equity tranche It should be noted that showing the fair spread for an equity tranche (here: 1194
similar to the other tranches in
order to to make them
bp) does not comply with the quotation for this piece in practice as stated before
comparable (always a constant premium of 500 bp plus the quoted upfront payment in terms
of the notional amount). We treat the equity tranche the same way as the other
tranches to make them comparable. Nevertheless, the special quotation
mechanism and fixing of the premium leg for the equity tranche does not
substantially change the risk characteristics of it.
Depicting the impact on fair However, we will not only focus on how the present value of a fixed contract
spread levels is an alternative
view that offers clear and
changes due to a variation of risk factors. As an alternative view, we always
favored hedge gauges depict the impact of changes to fair spread levels. This approach enables us to
calculate sensitivities in terms of basis points, which provides clear and favored
gauges when looking at the market behavior itself. For example, one might be
interested in the magnitude of a spread change for a tranche product provided
that the spread of the underlying index increases by 1 bp.


October 5, 2004
SENSITIVITY TOWARDS A MUTUAL SPREAD MOVEMENT

How do DJ iTraxx tranches At first, we will have a look at how a mutual spread widening or tightening
react to a mutual spread
widening or tightening?
affects (a) the present value of already existing tranches that carry fixed
spread levels based on the table above and (b) the fair spread levels of such
tranches in general. Just remember that individual spread changes are not
within the scope of the used HLPGC model. That is why we focus on equal
spread changes across all names of the underlying index. In the charts
below, we show the impact of an overall spread change in the range of 10 to
100 bp.
HOW DJ ITRAXX TRANCHES REACT TO A MUTUAL SPREAD MOVEMENT

60% 2500
Equity Equity
BBB
present value change (in % of the notional)
BBB
40% AAA
2000 AAA
Junior
Junior
Senior
fair spread level (in bp)
20% Senior
1500
0%
1000
-20%
500
-40%
-60% 0
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
mutual spread level (in bp) mutual spread level (in bp)
The way how present values As one would expect, an increasing spread of the underlying index leads to
and fair spreads react is as
expected, but differs across
decreasing present values of already existing contracts and rising fair spread
tranches levels for tranched products based on the index. However, the magnitude on
present values and spreads differs across tranches. The equity tranche shows
the highest sensitivity towards spread changes in the underlying index.
Apparently this is not surprising, because the first loss tranche incorporates
additional default risk to the largest extent, followed by higher tranches in turn.
The “spread delta” measures The right chart above enables us to calculate sensitivities in terms of basis
the sensitivity in terms of basis
points
points simply by extracting the slope of each curve. The slope of such a curve at
a certain point (we use a starting point of 60 bp for the underlying index) defines
the “spread delta”, which is given by
change in fair spread level of tranche i [in bp]
∆ i,spread = .
change in the spread of the underlying index [in bp]
Please note that this gauge is nondimensional owing to the same scale unit in
the numerator and the denominator. We computed the spread deltas for each DJ
iTraxx Europe tranche based on an initial spread level of 60 bp and obtained:
SPREAD SENSITIVITIES FOR DJ ITRAXX TRANCHES
Tranche Equity BBB AAA Junior Senior

Losses 0 – 3% 3 – 6% 6 – 9% 9 – 12% 12 – 22%
Spread delta 23.9 10.1 5.7 3.4 1.3
Spread PV -0.80 -0.46 -0.27 -0.16 -0.06
delta [%/bp]
Hedge ratio 16.6 9.5 5.6 3.4 1.3


October 5, 2004
Be careful when using spread As we can see, the equity tranche exhibits a spread delta of 23.9. This means
deltas for hedging purposes
that a 1 bp spread level increase in the underlying index leads to a spread
blowout of 23.9 bp for the first loss piece. This demonstrates the leveraged
exposure of tranched products in comparison to plain vanilla index investments.
But be careful! A delta of 23.9 does not mean that one needs a notional amount
in the underlying index that is 23.9 times the notional amount of the tranche in
order to eliminate the risk of spread changes. Three reasons for this:
– Eliminating the risk of spread changes requires that the present value of the
hedged item and the hedging instrument move in line, and not their fair
spread levels! Usually, the impact on present values resulting from a 1 bp
spread change (often referred to as “Sprd DV01”) is different for an index
swap and for a tranche based on this index. Thus, the “spread delta” is
inappropriate when building up a hedging relationship.
– The charts above already showed that the slope of the curves is not constant.
Different spread levels in the underlying index are accompanied by different
delta values, which means that the notional amount of the hedging instrument
regularly has to be adjusted according to the current delta values.
– The “spread delta” concept deals with the fair spread levels for index
tranches. This is the wrong approach for an existing position featuring a fixed
spread level, whereas the change in present value should be in the limelight
when tracking a position.
A few modifications are In order to solve the hedging problem, a few modifications to our spread delta
necessary to obtain an
adequate hedge ratio
concept have to be implemented. First of all, we have to define a delta measure
that involves present value changes for a contract with a fixed spread level. For
these purposes, we use the slope of the left chart above, and call it “spread PV
delta”. It is defined by
change in present value of tranche i [in fractions of the notional ]
∆PV
i, spread = .
Please note that this measure is independent from the size of the respective
contract, because the present value change in the numerator refers to the
nominal amount. Furthermore, we need to know how the underlying index swap
reacts with regard to a spread level change. Applying the HLPGC model to a
virtual “0% to 100% tranche”, which corresponds with a regular index swap
contract, solves this problem. Now, we can calculate our “spread PV delta”
above for this “special” tranche:
change in present value of the swap [in fractions of the notional ]
∆PV
swap , spread =
In our example we had a spread PV delta for the index swap of about –0.048%.
Thus, if the spread level increases by 1 bp, the present value of an index swap
contract (long credit risk) declines about EUR 4,830 given a notional amount of
EUR 10 mn. Consequently, we derive the sought-after hedge ratio by computing
the quotient of both spread PV deltas:
∆PV change in present value of tranche i

hedge ratio = =
i , spread
∆PV
swap , spread change in present value of the swap
Hedge ratios show a slightly We added the spread PV deltas for DJ iTraxx tranches and corresponding hedge
different risk profile than
spread deltas
ratios (dividing spread PV deltas by –0.048%) to our table above. In general,


October 5, 2004
hedge ratios do not coincide with spread deltas, particularly for tranches with
lower subordination (equity and BBB tranche). However, our obtained spread
deltas offer a good approximation for tranches with higher subordination.
Neutrality towards spread According to the table, an equity tranche investor who wants to eliminate the
changes raises additional issues
like ...
risk of spread changes, has to build up a contrarian index swap position with a
notional amount that is 16.6 times the size of the tranche contract.
Unfortunately, this raises additional issues that have to be taken into account:
... lack of carry-neutrality, ... – Apparently, the “hedged” position is not carry-neutral! The premium spread
of the equity tranche (1194 bp) is not totally offset by the hedging instrument
(16.6 times a spread of 60 bp = 996 bp).
... lack of neutrality towards – Even though spread neutrality can be assumed, there is a lack of neutrality
default risk, ...
towards default risk! Take a look at the amount that dwindles given a single
default within the DJ iTraxx Europe index (suppose a recovery rate of 40% as
assumed for our pricing model):
⋅ (1 − 0.4) ⋅
1 1
Equity tranche: ⋅ 1 = 16.0%
125 3% − 0%
⋅ (1 − 0.4) ⋅
1 1
Index swap hedge: ⋅ 16.6 ≈ 8.0%
125 100% − 0%
(first term: proportion of a single name within the index; second term: loss
given default; third term: allowance for the thickness of the tranche;
fourth term: notional multiplier)
... and unstable deltas and – As already mentioned, the hedge position has to be adjusted since spread PV
hedge ratios
delta and hedge ratios are not constant particularly with regard to spread
level changes of the underlying index. However, this phenomenon is not
unusual when trying to manage the risk of non-linear contracts, as are index
tranches.
The non-linearity of our The latter is evident on closer examination of the sensitivity graphs above, in
sensitivity graphs can be
clarified via delta curves
which the pronounced non-linearity of the curves is recognizable. In the
following charts, we clarify this aspect by drawing delta PV spreads and delta
spreads, respectively, against the mutual spread level:
SPREAD PV DELTAS AND SPREAD DELTAS SUBJECT TO DIFFERENT MUTUAL SPREAD LEVELS
0.0% 30
25
-0.5% Equity
20 BBB
spread PV delta
AAA
spread delta
-1.0% 15 Junior
Senior
Equity
BBB 10
-1.5% AAA
Junior
5
Senior
-2.0% 0
10 20 30 40 50 60 70 80 90 100 10 20 30 40 50 60 70 80 90 100
mutual spread level (in bp) mutual spread level (in bp)


October 5, 2004
The slope of spread (PV) delta In addition to the fact that deltas change when altering the mutual spread level,
curves is substantially
determined by the level of
we see different curve progressions for each tranche. The equity tranche shows
subordination the greatest change in delta (often referred to as “gamma”) combined with a
decreasing spread sensitivity for higher spread levels. For the record, the DJ
iTraxx Europe BBB tranche achieves its maximal spread PV delta (in absolute
terms) for a spread level of about 48 bp. However, the remainder (the other
subordinated tranches) exhibit a rising spread sensitivity for higher spread
levels. In a nutshell, the possible “gamma” problem has to be taken into account,
particularly for tranches with a low subordination level (especially first loss
tranches).
SENSITIVITY TOWARDS THE CORRELATION PARAMETER

HLPGC’s correlation parameter After delving into the subject of how mutual spread movements affect the
is a rough surrogate for default
correlation in reality, …
pricing of DJ iTraxx tranches, we now continue with the impact of changes
in the correlation parameter for the HLPGC model. Initially, we would like to
emphasize the already discussed issue that the correlation parameter could
only be understood as a rough surrogate for default correlation in reality.
… but gives us an idea how In practice, this gauge is rather used for quotation purposes than for pricing, as
correlation affects tranched
products
we already pointed out (please refer to the concept of “implied correlation”).
Nevertheless, varying this parameter within the HLPGC model gives us an idea
of how DJ iTraxx tranches react concerning correlation risk.
Increasing correlation leads to We already elaborated on how the correlation parameter changes the pool loss
a change in loss allocation with
respect to tranches while
distribution for the underlying index. While the expected loss of the pool loss
maintaining overall expected distribution remains unchanged when varying the correlation parameter, the
loss shape of distribution alters. Increasing correlation makes the tails of the loss
distribution fatter. Settings where no or many defaults occur become more
likely. Although the expected loss of the pool (the whole index) remains
unchanged, the loss allocation with respect to tranches varies.
HOW DJ ITRAXX TRANCHES REACT TO A CHANGE IN CORRELATION

0.6 4000
Equity Equity
BBB
0.4 BBB
present value change (in % of notional)
AAA
Junior 3000 AAA
Senior Junior
0.2
Senior
0 2000
-0.2
1000
-0.4
-0.6
0
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
correlation parameter correlation parameter
Changes for spreads/present As an increasing correlation parameter involves fatter tails of the pool loss
values are less clear for
mezzanine tranches
distribution, one would expect the fair spread of the equity tranche to decline
due to an increased probability of zero losses. On the other side, the fair spread
level for the senior tranche is expected to rise because heaped default scenarios
become more likely. Both charts above confirm that the equity tranche shows


October 5, 2004
the highest sensitivity. However, effects for mezzanine tranches are less
intuitive. The charts exhibit intervals with rising spreads and others involving
declining spreads.
SENSITIVITY TOWARDS THE RECOVERY RATE

The expected recovery rate is a Eventually, an analysis of how a change of the uniform recovery rate affects
risk factor, though it is usually
taken for granted
DJ iTraxx tranches is still missing. The market convention for recovery
rates (40%) is usually taken for granted, while reality often proves to be
different. Hence, investors should be aware of the fact that the expected
recovery rate is a market risk factor that is expensive to hedge (e.g. via
Digital Default Swaps).
As before, we depict the In this paragraph we outline how changes in expectations affect the present
marked-to-market change for
already existing tranches and
values of already existing DJ iTraxx tranches and their fair spread levels in the
their fair spread levels same manner we did before. The following charts impressively show that
different recovery rates (0%-100%) have a substantial pricing impact with the
highest sensitivity for the equity tranche. It is needless to say that tranche
spreads converge towards zero, if the recovery rate draws near 100%.
HOW DJ ITRAXX TRANCHES REACT TO A CHANGE IN THE EXPECTED RECOVERY RATE

40% 2500
Equity Equity
BBB BBB
present value change (in % of notional)
30% AAA AAA

2000
Junior
Junior
Senior
20% Senior
1500
10%
1000
0%
500
-10%
-20% 0
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
recovery rate recovery rate
Reducing the recovery rate ... In order to retrace the effect of a change in the expected recovery rate, let us
consider the BBB tranche (3-6% tranche). Given a decline from 40% to 30%, we
have to differentiate between the following effects:
... leads to a drop in the – While the contractual subordination level of the BBB tranche is fixed to a
effective subordination level of
mezzanine tranches ...
fraction of 3% of the notional amount, the effective subordination level,
measured in terms of necessary default cases until the prior tranches (here:
the equity tranche) erode, increases. A recovery rate of 40% implies a loss of
60% × 1/125 = 0.48% per default for the DJ iTraxx Benchmark Index. Thus, 7
defaults are needed until the equity tranche (0-3%) is exhausted and the BBB
tranche is affected. However, if we assume a recovery rate of 30% (implying a
loss of 0.56% per default), only 6 defaults would be needed.
... and shifts the pool loss – Secondly, reducing the recovery rate from 40% to 30% also means a jump in
distribution owing to a higher
expected loss for the index
the uniform spread level for the index constituents from 60 bp to 70 bp
provided a constant “clean” spread of 100 bp. This setting is accompanied by
a shift in the pool loss distribution owing to a higher expected loss for the DJ


October 5, 2004
iTraxx Benchmark Index.
A FEW WORDS CONCERNING THE PASSAGE OF TIME

Every investor should know As aforementioned, the elapse of time is no risk factor by itself because time
how his position evolves over
time concerning contractual
passes anyway. Nevertheless, it is necessary for an investor to know how his
cash flows and marked-to- position evolves over time, focusing on contractual cash flows and marked-
market changes. to-market changes. He should be aware of the following aspects assuming
all other risk factors to be constant:
– From a pure cash flow perspective, the protection seller receives quarterly
premium payments according to the initially stipulated spread level provided
no default occurs.
– Secondly, the pool loss distribution shifts to the left, attributable to the fact
that the expected loss declines through the passage of time. Less time to
maturity means a lower probability of experiencing enough defaults that
erode the tranches.
– In addition to the latter, a diminishing probability of (additional) defaults/
losses due to the elapse of time has the following consequence. While the
equity tranche still suffers from a potential danger of defaults, the chance of a
loss (default risk) for subordinated tranches ebbs away. The closer we get to
the maturity, the more of the (remaining) default risk has to be borne by the
equity tranche and not by subordinated tranches.


October 5, 2004
TRADING IDEAS & RELATIVE VALUE OPPORTUNITIES

Basic trading ideas The DJ iTraxx universe offers the opportunity to implement various trading
strategies, including plain-vanilla intra-DJ iTraxx trades (sector versus
sector) as well as more sophisticated cross-asset-class strategies, inter-
basket trades (DJ iTraxx versus default baskets) and even CPPI strategies.
Without claiming completeness, we point out the basic thinking of
implementing and analyzing the payoff-structures of such trading
strategies.
CORE-SATELLITE STRATEGIES:
COMBINATION OF DJ ITRAXX & SINGLE NAME POSITIONS
The DJ iTraxx is a perfect tool We see the establishment of a liquid credit index as a perfect tool for
for core-satellite strategies …
smaller clients to implement core-satellite strategies. A core-satellite
investment strategy in the credit universe includes building up exposure to a
market by buying a well-diversified index (the core investment), while
reducing/increasing idiosyncratic/sector risks (satellites). This is an
attractive strategy especially for smaller clients who do not have the support
of a crowd of analysts, which analyze every single name in the universe.
… also supported by the The implementation of this approach is very simple due to the equal-weighted
equally-weighted nature of the
index
nature of DJ iTraxx indices. The only restriction could be an investment volume
that is too low to hedge for single index constituents. Assuming an EUR 5 mn
long position in the DJ iTraxx implies that the investor carries a single name
exposure of EUR 40,000. This is not a sufficient size for a CDS contract. This is
even the case for a sector hedge. An investor who wants to hedge against sector-
specific risk in automobiles (which amounts to 8%) of the DJ iTraxx Europe
index, has to buy protection on a notional of EUR 400,000. Besides the bid-offer
problematic (which prevents a theoretically perfect hedge), the price for a small
notional may differ from the price for tradable sizes.
SINGLE NAME VERSUS SECTOR & SECTOR VERSUS INDEX

80 24 60 12
iTRAXX Autos 5Y VW 5Y CDS Diff. VW - iTraxx Autos (RS) iTRAXX Europe 5Y iTRAXX Autos 5Y Diff. Autos - Europe (RS)
70 21 55
10
60 18 50
8
50 15 45
40 12 40 6
6/20/04 6/30/04 7/10/04 7/20/04 7/30/04 8/9/04 8/19/04 6/20/04 6/30/04 7/10/04 7/20/04 7/30/04 8/9/04 8/19/04
Source: HVB Global Markets Research, Bloomberg
PV-neutrality versus carry- Let’s assume an investor who wants to sell protection on the DJ iTraxx
neutrality
Automobiles 5Y index for a notional of EUR 10 mn. However, the investor has a
negative view on VW, which has a 10% stake within the index. A PV-neutral
overall position requires a EUR 1 mn long protection position in VW. We assume
a current spread of the DJ iTraxx Autos of 45 bp and 5Y protection for VW


October 5, 2004
trades at 80 bp. In case VW spreads widen by 10 bp, the DJ iTraxx position

would generate a loss of EUR 4,650, while the long protection position in VW
would generate a profit of EUR 4,515 (using the CDSW function in Bloomberg;
standard settings). However, this trade is not carry-neutral in general! An
investor has to pay 80 bp on EUR 1 mn, while he receives 45 bp on EUR 10 mn.
That said, the costs for hedging 1/10 of the index exposure does, in general, not
equal 10% of the amount received for the long investment in the DJ iTraxx. This
is clearly linked to the fact that there is dispersion (please refer to the section
“pricing and quotation”). Only in case all sector constituents trade with the same
CDS spread, the above-mentioned hedge would be carry-neutral in any case.
SECTOR VERSUS BENCHMARK

Top-down investors can easily In the example above, we point out hedging strategies on a single name
manage their sector allocation
…
basis. The incentive of such a single name hedge is to eliminate idiosyncratic
risk. For investors who implement rather a beta-managed top-down
approach to run their credit portfolios would use sub-indices to reduce
and/or build up additional exposure to single industries.
As all sub-indices of the DJ iTraxx Europe carry similar strike spread (please
refer to the index description section), the dispersion on an industry level is
much smaller than on a single-name basis. Hence, the divergence between PV-
and carry-neutrality is much smaller regarding benchmark-sector trades.
… using DJ iTraxx sector indices Beta-management on a sector-level includes the active weighting of single
industries, in case the investor expects out- and/or underperformance potential
of single sectors. A DJ iTraxx Europe investor who is bullish on telecom bonds
(20/125) and bearish on the automobile sector (10/125), could switch from the
latter to the former by buying protection on Automobiles in the amount of
10/125 of the notional, while selling protection on telecoms in the same amount.
Assuming all indices are trading at their initial strike spread of 45 bp, the trade
position is nearly (ignoring dispersion effects) PV- and carry neutral. In case the
auto sector widened by 10 bp, the overall index should approximately widen by
0.8 bp. Multiplied by notional, the PV impact on the overall position is nearly
zero. A spread tightening of the telecom sector by 10 bp would generate a gain
on the total position of 30/125 * 10 bp in spread terms. In terms of PV
(assuming an initial notional of EUR 10 mn), a 10 bp tightening (only) in
telecoms would generate a profit of EUR 3,714 in the former Auto position
(notional of EUR 800,000), a profit of EUR 7,463 in the DJ iTraxx Europe
position (10 mn), while the long protection position in the auto sector remains
unaffected.
SECTOR VERSUS SECTOR & SINGLE NAMES VERSUS SECTOR

Idiosyncratic risk could be The example showed in the previous section could also be implemented in a
hedged easily
sector/sector and sector/single name trade. Including single names makes
carry-neutrality hardly achievable, while the P&L of sector/sector trades
without a DJ iTraxx Europe position closely follow the mechanism shown
above.


October 5, 2004
DJ ITRAXX VERSUS IBOXX

The DJ iTraxx is a appropriate Especially for hedging purposes, the DJ iTraxx is an attractive alternative
hedging instrument for iBoxx
investors …
for real money accounts to reduce net exposure to credits. While the
tracking error is negligible on a total return basis, roll-over costs & marked-
to-market valuations must be taken into account. Correlation between cash
& CDS markets should be, in general, relatively high, arguing for the DJ
iTraxx as a hedging instrument for iBoxx investors. But take good care of
the basis!
… although the number of As we already pointed out in our description section, the DJ iTraxx Corporates
constituents differs
significantly …
index is the appropriate hedging instrument for iBoxx investors. We also
mentioned that we view the duration-weighted character of the DJ iTraxx Corp
as somewhat problematic. Despite the fact that the index members of both
indices differ significantly (DJ iTraxx 100 vs. iBoxx > 200 names), the tracking
error will be rather limited on a total return basis.
… and there are differences in We see the differences in the construction of the iBoxx and the DJ iTraxx as a
the construction of both
indices
minor problem for hedging activities. The major task for the implementation of
hedging strategies is to properly analyze swings in the basis. The basis is the
difference between the credit default swap spread and the spread for cash
bonds, which could vary significantly over time. In other words, spread swings
in the iBoxx are not perfectly correlated to spread swings in the DJ iTraxx.
TAKE GOOD CARE OF THE BASIS!

60 -5.0 60 6.0
iBoxx non-financials - ASW spread (LS)
iTraxx non-financials (LS)
iBoxx - iTraxx (RS)
55 -7.5 55 4.0
iBoxx corporate - ASW spread (LS)
iTraxx Corp (LS)
iBoxx - iTraxx (RS)
50 -10.0 50 2.0
45 -12.5 45 0.0
40 -15.0 40 -2.0
06/21/04 07/01/04 07/11/04 07/21/04 07/31/04 08/10/04 08/20/04 08/30/04 06/21/04 07/01/04 07/11/04 07/21/04 07/31/04 08/10/04 08/20/04 08/30/04
Source: iBoxx, Bloomberg, HVB Global Markets Research
But: take care of the basis Although CDS and cash bond spreads are primarily driven by credit risk at a
first glance, there are a couple of fundamental and technical reasons for the
deviation between CDS quotes and cash bond spreads. In the following, we
analyze different factors which influence the default swap basis, which is
the difference between the CDS level and the asset swapped cash bond
spread.


October 5, 2004
BASIS DRIVERS
Impact Basis effect Comment

Counterpart risk ↓ Additional risk in bilateral CDS contracts
Funding issues ↓ Above Libor funding/ unfunded nature of
CDS/ funding risk
Market structure ↓ Short positions implemented in CDS markets
Volatility/Convertible trading ↑ Convertible players hedge credit risk via CDS
Market sentiment ↑↓ Front-running nature of default swaps
Cheapest-to-delivery option ↑ Delivery option favors CDS
Repo anomalies ↑ Repo-optionality of the cash investors
Profit realization ↑ “Offsetting” trade: Investor remains long
CPN step-ups ↓ Step-up features favors cash investors
Technical default risk ↑ Different default definitions for cash and CDS
CROSS-ASSET-CLASS: DJ ITRAXX VERSUS EQUITIES & SOVEREIGNS

From a macro perspective … In this section, we focus on the question how to structure a risk & return
optimized portfolio, including credits, stocks and safe havens. We explain
how credits reduce diversification costs in an overall portfolio (as is the case
for umbrella funds) and state that credits add value not only in a modest
growth scenario. Against this background, as well-diversified credit index is
a perfect tool as idiosyncratic risk is largely eliminated.
… optimizing the asset mix Regarding global portfolios, which include a broad range of asset classes,
requires an analysis of single
drivers for every asset class …
optimizing the asset mix requires an analysis of single drivers for every asset
class and correlation patterns between all assets / asset classes. While there are
a lot of possible input factors which could be used to feed factor models (also
known as Arbitrage Pricing Theory), interest rates are in the limelight for
umbrella funds.
… and correlation patterns Focusing on bonds & stocks, the traditional mechanism is that declining interest
rates go hand in hand with rising stock prices. The two major arguments are:
first, declining interest rates reduce refinancing costs and support capital
expenditures, boosting future company profits, which lead to a higher share
price. Second, declining interest rates will trigger allocation shifts from
government bonds into equities. This was (and probably still is) the mainstream
in most financial-economics textbooks. However, the problem is that this
mechanism has been suspended since the bust of the equity bubble in March
2000. This is most obviously reflected in the German DAX index & 5Y Bund
yields and the Bund Future, respectively. From 1990 until March 2000, there
was an inverse relationship between share prices and interest rates, which fits
with common theory. However, since March 2002 (all-time high in the DAX
index at 8064 on 7 March 2000), the picture changed dramatically. Interest
rates and share prices show a high positive correlation in the last 3½ years!
Does the monetary mechanism Does the monetary mechanism still work? In our view, it does. However, the
still work?
world is more complicated. The fact is that the sensitivity of economic growth to
interest rates may evolve over time due to structural changes but also due to
micro-fundamental developments (like overwhelming indebtedness of the
private sector or banks which have to repair their balance sheets and could not
continue their intermediary function). Without going into detail, a major impact
from evolving correlation for financial markets is the fact that the length of the
time series is decisive for the outcome of any correlation analysis. This is the
reason why we recommend investors to focus on scenario analysis rather than


October 5, 2004
on historical data.
Credits make the world for Besides this problem, which is inherent in all quantitative approaches, based on
umbrella funds shinier …
historical data, the good news is that there is an asset class, which makes the
world for umbrella funds happier. What mysterious asset class is this? Nothing
more than credits!
… given their attractive risk & Credits have a very attractive risk & return profile due to their three return
return profile
components credit return, curve return and accrued income. While the former is
positively correlated with equity returns (via implied volatility), the latter are the
fixed-income like performance components. What we call the offsetting process
is the fact that – on a total return basis – credits are the real safe havens in the
fixed income world as the credit & the curve return are negatively correlated. In
case of a bullish market sentiment, the curve return is negative due to rising
interest rates, while the credit return will be positive on the back of spread
tightening and vice versa.
Single return components of Hence, in a portfolio context, single return components of credits are negatively
credits are negatively
correlated
correlated to the other asset classes, stocks and government bonds. Negative
correlation is pure gold for portfolio managers and, in addition, makes hedging
much easier. Hence, credits could be used as a substitute for a combined equity
and safe-haven portfolio to a certain extent. This is especially the case for sub-
investment grade issues as the “equity-portion” of high yield bonds exceeds that
of investment grade credits. Having said this, credits have a positive impact on
the overall diversification process as they add a significant portion of
diversification gains in a portfolio. Moreover, adding credits in an umbrella fund
allow fund managers to reduce diversification costs. According to academic
research, the optimal diversification in an equity portfolio is at around 13%.
Further diversification will reduce the unsystematic risk component but the
related costs of further diversification overcompensate for the diversification
gain. Using credits is a simple & favorable way to reduce diversification costs
while keeping the diversification gains unchanged. Moreover, we can argue in a
Merton world about comparing the risk & return profile of credits with a short
put position. This can be directly used by a portfolio manager if he implements
option strategies for hedging reasons or to optimize the portfolio structure.
Credits traditionally offer value Despite the current tight valuation levels, credits traditionally offer value in a
in a modest growth
environment
modest growth environment. This is even more the case in global portfolios, as
we stated above. Hence, credits should always have a weighting above zero in
an umbrella fund. From a pure theoretical standpoint, credits can substitute the
portfolio’s risk & return contribution of safe havens due to the fact that the
return components of government bonds could also be found in credits. Hence,
under the assumption that every maturity profile could be displayed, investors
can construct a portfolio including equities & credits, which has exactly the same
risk & return profile as a portfolio including government debt and equity. Due to
the equity-related risk component of credits (the spread return), the correlation
between equities and credits is higher than between the former and
governments. This means that the equity share could be reduced in an
equity/credit portfolio compared to a governments/equity portfolio, which
generates the same risk & return profile.
The advantages of credits The advantages of credits are: 1) there may be arbitrage opportunities as there
is more than one alternative to reach a specific risk & return profile due to the
substitute character of credits; 2) diversification costs in the equity share could
be saved as credits already provide “natural” diversification effects. Last but not


October 5, 2004
least, in a CAPM world, a variable share of equities/governments means the

optimum on the capital market line is not fixed! Depending on the shape of the
utility function, this could cause a jump to a higher utility curve. However, this is
not a necessity!
DJ ITRAXX VERSUS DEFAULT BASKETS

Similarities and differences Synthetic CDO tranches and N-th-to-default (NtD) CDS baskets are similar
between CDO tranches and N-
th-to-default swaps.
financial products. In contrast to single name derivatives like CDS they are
based on a portfolio of credit-risky assets. For a portfolio manager, they
provide good protection against the risk of an adverse clustering of defaults
at lower costs than hedging all individual exposures with single name CDS.
A protection seller might achieve an attractive premium and gain leveraged
exposure to a portfolio of credits, which might be inaccessible otherwise.
For both derivative structures, default correlation is the name of the game
in order to model the payoff profile and the value. But for valuation
purposes, a superficial analysis of the structures may lead to wrong results,
since a CDO tranche is not just a bunch of NtD baskets. To be short: the NtD
CDS basket involves the probability of a certain number of defaults, while
for CDO tranches one needs the probability of an accumulated realized loss.
A simple example sheds some light on the differences.
Delta-neutrality … Regarding the mechanism of a first-to-default basket (FTD), the legal
construction as well as the trigger event are similar to a single Credit Default
Swap (for analyzing default baskets, we recommend to use the CDSN function on
Bloomberg). In case of a credit event, the protection seller pays
(1 – recovery rate) × notional amout of the defaulted issue
to the protection buyer (cash-settlement) or the protection buyer hands over
deliverable obligations and receives the notional amount (physical settlement).
Following the first default, the FTD-contract is terminated and the protection
buyer loses any further protection for the other basket constituents. As the
protection buyer is only insured against the first default, the basket spread is
below the sum of the CDS levels of all basket credits. A basket is attractive for
both counterparts. From the protection buyer’s standpoint, the protection fee is
cheaper compared to CDS on single names, while the protection seller gains
leveraged exposure to a basket of credits as it enables the investor to earn a
higher yield than any of the credits in the basket. In any case, the protection
seller has to closely monitor his delta & gamma positions, as these are highly
sensitive to price changes of the underlying credits and could be understood as
additional risk components.
… but huge gamma risks Besides the risk of default, the issuer of a default basket is exposed to spread
swings (as he carries a short position in credit risk), which could be hedged by
selling protection via CDS. In case the spread of one or more portfolio
constituent changes, the protection buyer’s initial position is over/under-hedged.
For the implementation of optimal hedging strategies, the hedge ratio needs to
be determined (i.e. which notional must be sold), and this requires knowing the
spread delta of the position. Besides the problem of determining portfolio
“greeks”, using historical data or modeling those stochastic processes, in case of
a default of one basket constituent, negative effects on the spread of the
remaining issues could cause a loss when the position is unwound. Another
possibility for an investor in a FTD to hedge against default is to buy protection


October 5, 2004
for a combination of NtD baskets (please refer to the following paragraph),

including similar credits. The problem of this strategy is to find counterparts
who are willing to sell protection on similar baskets.
Hedgers via long protection Besides the difficulty of determining sensitivities for a delta-hedge, hedgers via
also carry a long position in
gamma
long protection also carry a long position in gamma (sensitivity of the delta in
respect to spread changes). In case of spread widening of a basket credit, the
delta will rise and the basket is over-hedged. Hence, the hedger will try to sell
protection at wider levels. In case of narrowing spreads, the basket is under-
hedged and the hedger will buy back protection at lower levels. Therefore,
accelerating spread volatility favors the hedger and could help compensate for
the negative carry of a hedged FTD basket. However, the possibility of dynamic
hedging is overestimated as adjustment-trades to spread changes are limited
because of the (at least for the time being) lack of liquidity for individually
structured products and CDS on exotic names in secondary markets. Moreover,
the small deal-size of baskets (single-digit million area), dynamic hedging
requires to sell/buy credit risk in very small portions (let’s say EUR 20,000),
which is not realistic. By the way, maturity mismatch could be another problem
as CDS markets lose significant liquidity regarding non-standardized maturities.
CDO tranches and NtDs have Usually, CDO tranches and NtDs focus on different portfolio sizes. A typical NtD
different applications …
basket contains up to 10 reference entities, while the portfolio underlying a CDO
transaction might be larger, up to a few hundred credits (e.g. the recently
established tranched DJ iTraxx products). For our analysis, we assume a
portfolio of 10 single-name CDS contracts. For the sake of simplicity, we further
assume that all CDS are written on the same volume (i.e. 10% of the total
volume) and involve a uniform non-stochastic recovery rate of say 40%. These
assumptions are crucial for the following discussion, given their impact on the
pricing of CDOs and default baskets.
… but have some interesting The similarity of both structures becomes obvious if we make clear that the 1st-
similarities
to-default CDS is related to the first 10% volume of the portfolio, the 2nd to
10 – 20% volume and so on. Thus, a naive (or rule of thumb) approach would
suggest that a hypothetical 10 – 40% tranche of a CDO is the same as a portfolio
of a 2nd-, 3rd- and 4th-to-default CDS. If we knew the spread of the NtDs, we
could easily calculate the fair spread of the tranche by taking the average spread
of the related NtDs.
Recovery matters … Why is this approach wrong? The first and most obvious reason is recovery.
Tranche derivative products are usually defined in such a way that just the
realized losses eat into the tranches. The first default hits the 1st-to-default CDS,
but since there will be a recovery payment, the loss of the first default will not
completely extinguish 10% of the outstanding portfolio notional. So there will be
something left from the 0%-10% tranche after the first default. One difference
lies in the fact that in the case of a default, the 1st-to-default investor has to pay
the notional, receives the recovery and the game is over. While the tranche
investor suffers a decline in the outstanding balance, he still receives the
premium on the rest of the tranche. Clearly, we could adjust the tranches to
cover this problem. In our example, we would define a 0%-6% tranche. But then
we will face the next problem: the uncertainty of the recovery payment. Most
probably, it will not be exactly 40%, but will deviate from the expected value
quite substantially. But since we want to stick to our example, we suppress all
these recovery problems with a brave assumption: zero recovery.


October 5, 2004
… as well as different default We admit that this is unrealistic but it helps explain a second problem. We
impact on premium payments
compare the portfolio of a 2nd-, 3rd- and 4th-to default CDS with the 10 – 40%
tranche. Now let there be two defaults which strike our portfolio and both
derivative structures suffer. Since there will be no recovery, both loose a third of
their volume. What is the difference between the portfolio of NtDs and the
tranche? The CDS investor says goodbye to his 2nd-to-default swap. On his
remaining portfolio, he now receives the average of the spread of 3rd- and the
4th-to-default swap, which will certainly be lower than the average spread just
before the default, since the NtD spread decreases with an increasing number of
assets. Additionally, he earns this reduced spread on a reduced outstanding
notional. He gets punished on the spread and on the volume. Compared to that,
the situation for the tranche investor seems to be more comfortable. He just
faces a loss on the tranche volume. The spread he receives on the remaining
volume is fixed.
Same default legs, but We conclude that the investment strategies have different payments in the
different premuim legs
premium leg but the same realization in the default leg. These different
payments will be reflected in the pricing. Thus, for pricing a CDO tranche we
cannot just take the average price of the corresponding NtDs. From a modeling
point of view, the NtD involves modeling the number of defaults, while for the
CDO tranches the accumulated realized loss is of high importance.
No rule of thumb While there is no rule of thumb with respect to the spread difference between a
default basket portfolio and a tranche-product (including the same credits), in
our example the spread on the tranche is lower than the average spread of the
basket portfolio.
PLAYING THE STEEPNESS OF THE DJ ITRAXX CURVE

A flattening trade can easily be As there are two maturities in the DJ iTraxx universe (5Y and 10Y),
implemented with 5Y and 10Y
DJ iTraxx swaps
investors can easily implement their view towards the average steepness of
credit curves. In the following example, we anticipate the flattening of credit
curves, with the short end underperforming the long end (e.g. due to rising
risk aversion). In the following, we depict the mechanics of a possible
flattener trade utilizing the 5Y and the 10Y DJ iTraxx Benchmark swaps.
Important criteria for curve Before introducing a possible trading constellation, we should be aware of which
trade
criteria could be important in such a context. The trade should not involve a
downside risk in case of default, which means that the long position in credit
risk (selling protection) shall not exceed the short position (buying protection).
Additionally, the trade should have no or a positive carry in order to avoid any
negative cash flows. If possible, the P&L impact of a parallel shift of the DJ
iTraxx credit curve should be minimized. However, accomplishing all three
objectives with only two contract types is usually not be possible due to a
missing degree of freedom. This is why we start with the first two goals and
assess the third one afterwards.
A flattening trade may Against this background, a possible trading structure may look like this:
encompass buying 5Y
protection and selling 5Y – Buy protection in the 5Y DJ iTraxx Benchmark swap at 37.5 bp (ask quote).
protection (carry neutral)
– Sell protection in the 10Y DJ iTraxx Benchmark swap at 50 bp (bid quote).
The notional of this contract is adjusted to carry neutrality, amounting to
37.5/50 = 0.75 times that of the first contract.
The concept behind the trade is to gain from a spread widening at the short end


October 5, 2004
(5Y) while anticipating rough stability in the 10Y bracket. Even though a 10Y
credit index swap shows a longer duration and therefore a higher sensitivity to
spread changes as the 5Y contract, the maturity mismatch is to a large extent
offset by the reduced notional amount (0.75 vs. the proportion of duration of
about 0.59). In the following table we point out the structure of the trade:
DIFFERENT SCENARIOS FOR OUR TRADING RECOMMENDATION*
Current Trade Scenario 1 Scenario 2 Scenario3 Scenario 4 Scenario 5

spreads position upward shift downw. shift 20% flatten. 30% flatten. 40% flatten.
5Y DJ iTraxx swap 36.5 / 37.5 Buy protection 42.0 32.0 39.8 41.2 42.6
10Y DJ iTraxx swap 50.0 / 52.0 Sell protection 56.0 46.0 51.0 51.0 51.0
MTM 5Y swap +22,463.60 -27,455.50 +11,481.39 +18,470.07 +25,458.74
MTM 10Y swap -37,805.80 +25,203.84 -6,300.96 -6,300.96 -6,300.96
Total MTM -15,342.20 -2,251.66 +5,180.43 +12,169.11 +19,157.78
* Indicative levels from October 1, 2004. The calculation is based on a notional amount of 10 mn for the 5Y and 7.5 mn for the 10Y contract. The computation
of marked-to-markets (MTM) is subject to spread changes immediately after initiation of the trades. MTMs are calculated on mid levels.
Source: HVB Global Markets Research.
This trade generates a The first two scenarios correspond to settings where the spread curve performs
significant profit in case of a
flattening trend in the credit
a parallel shift upwards and downwards. The impact on the total PV is rather
spread curve for the DJ iTraxx limited due to the above-mentioned partial duration neutrality. Scenarios 3 to 5
Benchmark Index assume a decline of the credit curve steepness by 20%, 30%, and 40%,
respectively, leaving the 10Y level unchanged. In this case, our trade generates
significant profits, which can be realized by termination of both contracts.
SPREAD OPTIONS: SINGLE & COMPLEX STRATEGIES

DJ iTraxx Europe index acts The DJ iTraxx Europe index also acts as an underlying for spread options.
also as an underlying for
spread options
Options allow investors to implement simple hedging strategies (portfolio
insurance / protective put), and also more complex strategies (straddles,
strangles, butterflies, etc.). As pricing models have to incorporate the fact
that credit spreads are not normally distributed, a modified Black & Scholes
model extended by a stochastic jump process is an appropriate approach.
Spread options are available for the DJ iTraxx Europe index S2 March 10. There
are payer options (calls) and receiver options (puts), with strike prices of 30, 35,
40, 45, 50. The current expiration dates are scheduled on December 2004,
March 2005 and June 2005.
Hedging strategies & Hedging strategies for credit portfolios (portfolio insurance strategies) or long
positions in iTraxx indices (protective put) could be easily implemented. Trading
strategies include complex strategies like straddles, strangles, butterflies,
condors, etc. These strategies are well known from equity derivatives markets
and hence, we abstain from a detailed analysis.
We suggest to use a Black & The major problem is linked to the pricing of the option. An exact definition of
Scholes model extended by a
stochastic jump process
an investor’s risk positions (the Greeks: delta, gamma, vega and theta), require
an appropriate pricing model. A possible approach is to use the Black & Scholes
framework, extended by a stochastic jump process to cover the problem that
credit spread swings are not log-normally distributed. While the Black & Scholes
approach is widely accepted, the problem lies, in our view, in determining the
parameters for the jump process. “jump intensity” and “jump width”, the two
major parameters for a jump process vary significantly over time. Estimating
jump-processes depends heavily on the market environment. As long as there is


October 5, 2004
no standard model to price spread options, investors are exposed to so-called

model risk. An investor may calculate the theoretically fair price of the option
correctly, but all other market players use different models and calculate
different prices. In this case, P&L realization would be affected.
CPPI STRATEGIES INCLUDING DJ ITRAXX INDICES

An interesting payoff profile Given the well-diversified nature of the DJ iTraxx Europe, the index family
could be generated …
could be seen as an interesting risky asset within CPPI (Constant Proportion
Portfolio Insurance) strategies. In combination with risk-free instruments,
an interesting payoff profile could be generated.
… by using DJ iTraxx indices The CPPI was introduced in the late eighties for FI and for equity instruments,
within CPPI strategies
implementing a dynamical asset allocation over time. In a first step, the investor
sets a floor equal to the lowest acceptable value of the portfolio. This determines
the so-called cushion as the excess of the portfolio value over the floor, which
define the amount allocated to the risky asset by multiplying the cushion by a
predetermined multiple (dependent on the “riskyness” of the asset). Both the
floor and the multiple are functions of the investor’s risk tolerance and are
exogenous to the model, while the total amount allocated to the risky asset is
known as the exposure. The remaining funds are invested in the risk-free asset,
usually AAA-rated government bonds. The higher the multiple, the more the
investor will participate in a price increase of the risky asset, while a higher
multiple means that the portfolio will approach the floor faster in case of a
sustained price drop. As the cushion approaches zero, exposure approaches
zero too. This prevents the portfolio value from falling below the floor.
MECHANISM OF A CPPI STRATEGY INCLUDING THE DJ ITRAXX AS THE RISKY ASSET
Additional
Spread income
Risk-puffer
iTraxx
cushion
iTraxx
Risk-free asset
Risk-free asset
floor
T=0 T=1 T=2 time
Combining a risky and a risk- The major ingredients for a CPPI strategy are a risky and a risk-free asset.
less asset
Therefore, one may think to combine corporate credits and government bonds
in such a strategy. While this would lead to a relatively high multiple (as the
total return variance of credits is relatively low, i.e. compared to equities), a
well-diversified credit index reduces management costs as idiosyncratic risk is
already eliminated to a large extent. Dealing with company-specific risk is not
the major competence of a CPPI-manager.


October 5, 2004
OUTLOOK ON PRODUCT DEVELOPMENT

Strong growth of credit The strong growth of credit derivatives markets, which we think will persist
derivatives markets will be
accompanied by the ongoing
in the coming years, is accompanied by the ongoing development of new
development of new instruments. While “plain-vanilla” derivatives will experience rising
instruments liquidity, there are a couple of exotic derivatives which are rather tailor-
made instruments, which will be traded by only a few investors (hedge
funds, etc). The major trends in the iBoxx universe, in our opinion, will be
the introduction of different maturities. in line with single-name CDS
contracts (1Y, 2Y, 3Y, 5Y, 7Y, 10Y). Furthermore, we would highly
appreciate options on tranches, enabling investors to trade volatility of
correlation!
Future on the DJ iTraxx Europe, The most common instrument in the derivatives world will be introduced in the
…
iTraxx universe, too. The IIC plans a future on the DJ iTraxx Europe at the
beginning of 2005, without officially having announced many details yet. We
assume that the contractual standard of the future contract will be in line with
plain vanilla FI futures (Bund future & Co) traded at the Eurex. We expect an
iTraxx future contract will be highly appreciated by the investor base, as it
provides a high degree of liquidity and offers a simple way to gain exposure to
directional spread risk. Pricing would be rather straight forward, in line with
well-known above-mentioned future contracts, without any sophisticated
modeling needs.
… and different maturities on A milestone for credit investors would be the introduction of iTraxx indices
iTraxx indices will be the next
milestones
across the whole maturity curve (1Y up to 10Y contracts as it is the case for
single-name CDS). In case the iTraxx is priced through the whole curve, forward
contracts could be introduced. Forwards offer the opportunity for investors to
implement spread views for a specific time interval in the future and are crucial
for portfolio management as well as for asset-liability players. One may argue
that it is only a question of time until the iTraxx is priced across the whole curve
given the declining TTM for already launched contracts (the iBoxx March 2010
Series 1 will have a “maturity” of 7Y at the end of 2006). This is only a second-
best solution as the underlying universe may change dramatically given the
passage of time, which argues against a consistently priced curve. Assuming
that these iTraxx indices would also include tranches, a term structure of
correlation could be derived. Against this background, “playing the correlation
curve” might be a future trading strategy for correlation desks.
Although a couple of From a theoretical point of view, there is a broad range of highly interesting
“derivatives squared” are
thinkable, …
instruments, including floating-spread products or tranche options. But as there
is, in our view, only a small portion of the credit community which is able to
correctly price and trade such “derivatives squared”, we do not expect the
… the focus will remain on establishment of a liquid market. The focus will remain on liquid products,
liquid products
which serve hedging needs, trading activities and arbitrage opportunities for a
larger share of credit players.


October 5, 2004
HOT TOPICS
SYSTEMATIC RISK OF CREDIT DERIVATIVES

The other side of the coin The ongoing development of the DJ iTraxx index family will continue to
trigger increasing liquidity in the credit derivatives market. Increasing
liquidity leads to a more efficient market structure, lower transaction costs
(at least via reduced bid-ask spreads) and increasing transparency. Besides
these “old-school arguments” in regard to the advantages of credit
derivatives markets, there is also another side of the coin: Increasing
systematic risk!
Systematic risk arises through Systematic risk arises through leveraging (synthetic) exposure to underlying
leveraging
assets, namely if the notional amount of credit derivatives significantly exceeds
the outstanding amount of the underlying debt. Assume a company has EUR 5
bn debt outstanding, while CDS contracts have a notional amount of EUR 10 bn
(there is no natural or legal limit of the volume of OTC-traded derivative
contracts). As physical delivery for all outstanding CDS contracts is obviously not
possible, cash settlement must be agreed upon. This raises, at a first glance, no
serious problems. From a purely technical standpoint, the derivative is
completely decoupled from the underlying cash instrument. But, and this is
certain, there is a pricing link between both instruments. Exactly the same is
true for synthetic basket products, like CDS-based credit indices.
The derivative is the front- In theory, price swings in the underlying asset determine the price of the
running instrument
derivative. However, in practice it is often the other way around, where the
derivative is the front-running instrument. Besides the fact that market
anomalies in derivative markets could lead to a deviation of the cash price from
the underlying credit fundamental, any negative impact is primarily related to
the issuer rather than to investors.
Our major concern is linked to Our major concern is linked to the fact that credit exposure could be leveraged
the fact that credit exposure
could be leveraged through
through derivatives. Textbook-like arguments for the derivatives market focus
derivatives on the risk dispersion & transfer function of derivative instruments. As we often
mentioned, we think this is only one side of the coin. While credit risk can be
divided into its components, which can be traded separately, the incentive to
originate credit risk will increase! Despite the fact that this risk will be
distributed to many investors, reducing unsystematic risk, the overall amount of
originated systematic risk will rise. If the overall systematic risk is likely to rise,
this will lead to an increased fragility of the financial system as a whole! To
make a long story short: Credit derivatives – not suitable for kids!


October 5, 2004
THE DJ ITRAXX FAMILY FROM THE VIEWPOINT OF PORTFOLIO RISK

MANAGER AND HEDGERS
The DJ iTraxx is not the Credit risky portfolios are in general prone to three (interrelated) risk
philosopher’s stone for
hedging portfolio credit risk …
factors: individual default risk, spread risk and the risk of correlated
defaults. Risk management and hedging portfolio risk needs to deal with all
three topics. DJ iTraxx products can be useful for this purpose. However,
the DJ iTraxx is not the philosopher’s stone for hedging portfolio credit risk.
… as most portfolios will be Hedging default risk (i.e. effectively reduce the exposure to specific names) with
substantially different from DJ
iTraxx indices
DJ iTraxx is a difficult task because of the standardized portfolio structure in the
DJ iTraxx. Most portfolios will be substantially different from DJ iTraxx indices,
concerning both the selection of the individual constituents and their
corresponding portfolio weightings. Hence, buying protection on DJ iTraxx will
not remove default risk comprehensively. If the hedged portfolio is not very
similar to the DJ iTraxx, a customized transaction will most probably be more
suitable, although transaction costs will be somewhat higher in this case.
Perfectly hedging spread risks Hedging the spread risk component of a customer portfolio with the DJ iTraxx
…
might be attractive, even if the composition of the portfolio is different from the
DJ iTraxx. In case there is a satisfactory correlation between the index spread
and the spread in the portfolio, the index price changes will largely offset those
in the hedged portfolio, at least under normal market conditions. A word of
caution here as well: defaults can (and most probably will) trigger adverse
market movements, which can lead to the breakdown of well-established
correlation patterns, making the hedge inefficient.
… requires the analysis of There is also an upside to the latter point, which can be exploited for hedging
spillover effects
purposes. In case of a credit event of a major counterpart, spreads of related
entities will be negatively affected (spillover effect). The worst-case scenario is
that a single default triggers other defaults, known as default-contagion. Thus,
the DJ iTraxx spread may react to default, even if the defaulted entity is not
included in the index.
Swapping the concentration In addition to a pure hedging-approach (reducing exposure), diversification is
risk into the DJ iTraxx
another important topic in portfolio management. Against this background, the
DJ iTraxx family offers interesting opportunities. In case of concentration risk,
the risk manager can easily switch from single-names to a diversified portfolio:
buy protection on single names and sell protection on the DJ iTraxx. This
essentially means swapping the concentration risk into the DJ iTraxx, which is
already diversified (see also the relative value ideas below).


October 5, 2004
BLOOMBERG CREDIT DERIVATIVES FUNCTIONS

Bloomberg offers some For valuation purposes, Bloomberg offers some interesting analytic tools
interesting analytic tools
which simplify life for credit derivatives investors. The major Bloomberg
functions regarding credit analysis and derivatives valuation are listed in
the table below.
USEFUL BLOOMBERG ANALYTICS
Function Description
Credit Derivatives functions
CDSW Value credit default swaps of single issuers. Including implied default probabilities.
CDSD Set up custom curve defaults, valuing credit default swaps.
CDSN Analysis and creation of default basket. Can be used for DJ iTraxx FTD valuation.
NI CDRV News on credit derivatives.
Spread & Swap valuation
ASW Structure and value asset swaps. Hedging swap cash flows for any FI security
RVS Shows the history of spreads versus swaps of a single bond.
OAS1 Calculating option values and pure credit spreads for bonds with option features.
WS Current and historical global swap spreads.
SWPM Valuation of interest rate swaps and derivatives. Risk & horizon analysis for swaps.
IRSM Menu of interest rate and derivative function.
SWDF Interest rate swap curves
SSRC Monitors swap rates, forward rate agreements, swaption volatilities, cap and floor
volas and spot future strips.
FMC Yield curve analytics, incl. spreads and term structures for specific sectors/qualities.
Credit Quality functions
DDIS Maturity distribution of a selected issuer’s outstanding debt.
CRPR Current rating for a specific issuer and/or a specific fixed income security.
RATC Current and historical rating actions.
RATD Rating scales and definitions for various rating agencies


October 5, 2004
APPENDIX A: DERIVING THE EXPECTED LOSS FUNCTION

In this section, we present the derivation of the expected loss function by means
of the “homogeneous large portfolio gaussian copula” model (HLPGC). The
model is based on the Vasicek model (Vasicek 1987, 1991) on loan loss
probability distributions. In the derivation, we follow Schönbucher (2003).
Before we start, let’s recall what we are heading for: we want to derive a
r r r
( )
mathematical expression for the expected loss function ELba t , Ti , ρ , s , R . As
already mentioned, the “homogeneous large portfolio gaussian copula” (HLPGC)
model needs just three parameters, an average correlation, an average spread
and an average recovery rate. We can thus write ELba (t , Ti , ρ , s, R ) (note that
ρ , s, R are scalars and not vectors).
In the first step, we have to calculate the default probabilities for the individual
credits from the (dirty) spread s and the recovery rate R .
p (t , T ) = 1 − e − s ⋅(T −t )
s (1)
with s = (clean spread)
1− R
The term p(t , T ) is the probability that a default occurs between t and T (given
that no default occurred until t ). Consequently, the survival probability is given
by e − s ⋅(T −t ) . The term T − t is the time-to-maturity (a.k.a. the investment
horizon). Remember that using HLPGC assumes the same default probability
function for all underlyings.
The main idea of the HLPGC model is to model the default correlation using a
simple firm value approach. In the firm value models, which date back to Robert
Merton (1974), one assumes that the default risk of a firm can be modeled
through the dynamics of the firm’s assets. Once asset values fall below a certain
barrier, a default is triggered.
Homogeneous model The value of the firm’s assets at time T is denoted by Vn (T ) , which is modeled
as a normally distributed random variable. To incorporate the default risk, we
need the default barrier K (T ) that triggers the default. Since the default
probability is time-dependent, the default barrier is time-dependent too. As we
assume that the default probability for all credits is the same, all credits have
the same uniform default barrier K (T ) . This is why we call this model
homogeneous. In a more general approach, we would have company-specific
default barriers K n (T ) . The default condition is given by:
V j (T ) ≤ K (T ) ⇒ default (2)
In this equation, V j (T ) is a stochastic variable, while K (T ) is deterministic. For

the sake of completeness, we add that V j (T ) and K (T ) depend on the time-to-
maturity T − t , but we omit the t dependency in the expressions to facilitate
readability. The level of the default barrier K (T ) can be calculated by using a
simple idea. The difference between the initial value V j (t ) (note the t instead of
T ) and K (T ) is adjusted such that the probability of V j (T ) hitting K (T ) is the
same as the default probability p(t , T ) up to time T :
[ ]
Prob V j (T ) ≤ K (T ) = p(t ,T ) = 1 − e − s ⋅(T −t ) (3)


October 5, 2004
Since the absolute value of V j (T ) is not interesting, we set V j (t ) initially to zero.

Thus, we can use the inverse of the cumulated standard normal distribution
Φ −1 ( ⋅) – remember that we assume that V j (T ) is a normally distributed random
variable – to calculate K (T ) from p(t , T ) :
(
K (T ) = Φ −1 ( p(t , T )) = Φ −1 1 − e − s ⋅(T −t ) ) (4)
To illustrate this, we present a numerical example. We assume a flat credit

spread s of 100 bp and a recovery rate R of 40%. The clean spread s is
therefore about 167 bp. The default probability p(t = 0, T = 5) for an investment
horizon of 5 years is 8%. The corresponding default boundary K (T = 5) is –1.41.
This means that the asset process, which started at zero, hits the default barrier
of –1.41 within the next five years with a probability of 8%.
It is also interesting to note that we “inverted” the Merton model in this
approach. Usually, the Merton approach aims at deriving the credit spread from
asset volatility. The default barrier is estimated from balance sheet data, such as
the debt to equity ratio. Here, asset volatility is standardized to 1 (since credit
risk for all underlying is assumed to be equal) and the default boundary is
derived from spread data and not vice versa.
In the next step, we have to model the default dependency structure. This is
done by simply assuming that all asset values are correlated with a single
market variable. Hence, it is a single-factor model. You can view this market
factor as an abstract measure of the state of the business cycle.
V j (T ) = ρ ⋅ M + 1 − ρ ⋅ ε j ∀j = 1,2,K, n with 0 ≤ ρ ≤ 1 (5)
Gaussian copula Here ρ is the correlation of firms’ asset values with the market factor M . The
variable ε j is the idiosyncratic, firm-specific default risk of each credit j in the
portfolio. Although all credits have the same default probability and asset values
are correlated with a common risk factor, the individual defaults are
independent, conditional on M , since ε j ’s are independent. Since we
subsequently assume that M and all ε j ’s are normally (gaussian) distributed,
we refer to the dependency structure as a gaussian copula.
If the correlation is zero ( ρ = 0 ), there is no effect from the market factor. The
portfolio risk is completely driven by individual credit risks. If the correlation is
one ( ρ = 1 ), default risk of all companies is solely driven by the state of the
business cycle, firm-specific factors do not play any role then.
Equation (5) says that the asset values V j (T ) are conditional on the systematic
risk factor M , driven by the independent normally distributed random variables
ε j . By using the default trigger definition of equation (2) we obtain
ρ ⋅ M + 1 − ρ ⋅ ε j ≤ K (T ) or
K (T ) − ρ ⋅ M (6)
εj ≤
1− ρ
From equation (5) we can derive the probability of a default conditional on the
level of the risk factor M .


October 5, 2004
 K (T ) − ρ ⋅ M 
[
p j (t ,T | M ) = Prob V j (T ) ≤ K (T ) M = Probε j ≤] M
 1− ρ 
 K (T ) − ρ ⋅ M 
= Φ  (7)
 1 − ρ 
 
= Φ
( )
 Φ −1 1 − e −s ⋅(T −t ) − ρ ⋅ M 

 1 − ρ 
 
Φ( ⋅) is the cumulated standard normal distribution function. The individual

conditional default probability p j (t ,T | M ) is the probability that the firm’s asset
value V j (T ) is below the default barrier K (T ) at time T , conditional on the level
of the market risk factor M .
Large portfolio Now we have individual default probabilities conditional on M . The next step is
to join them together to a portfolio default distribution. Here our “large”
assumption comes into play. In a large portfolio with credits of the same size,
the fraction of defaulted credits is given by the average of the individual default
probabilities, conditional on the state of the business cycle.
 K (T ) − ρ ⋅ M 
d PF (t , T | M ) = Φ  (8)
 1− ρ 
 
The mathematical expression of d PF (t , T | M ) and the individual default

probability p j (t ,T | M ) are exactly the same, but the interpretation is different.
The term p j (t ,T | M ) is the probability for firm j to default up to time T ,
conditional on the state of the economy M . d PF (t , T | M ) is not viewed as a
probability, but rather as the default fraction of the portfolio up to time T ,
conditional on the state of the economy M .
The loss fraction can be calculated by introducing the recovery rate
l PF (t , T | M ) = (1 − R ) ⋅ d PF (t , T | M ) (9)
As an illustration, we plot the l PF (t , T | M ) as a function of the level of the market

factor M . In the left chart below, we show the curve for different values of the
correlation parameter ρ . All curves approach asymptotically the level of (1 − R ) ,
which is 50% in the current example. The larger the parameter ρ , the faster it
approaches its asymptotic value. In the right chart below, we show the
dependency on the spread value of the underlying index. Changes in the spread
shift the curve on the x-axis to a greater or lesser extent.


October 5, 2004
SHAPE OF THE LOSS FUNCTION FOR VARIOUS PARAMETER VALUES

FOR SELECTED CORRELATION VALUES FOR SELECTED SPREAD VALUES
60% 60%
50% 10% 50% 0.1%

conditional default fraction
conditional default fraction

20% 0.5%
40% 30% 40% 1.0%
40% 5.0%
50% 10.0%
30% 30%
20% 20%
10% 10%
0% 0%
-6 -4 -2 0 2 4 -6 -4 -2 0 2 4
market factor M market factor M
The loss fraction l PF (t , T | M ) is the key quantity to calculate the expected loss of
a specific tranche. We can proceed in two different ways.
1) From l PF (t , T | M ) we can derive the unconditional loss probability. This
quantity gives the probability for a specified loss fraction and for a given
time horizon, e.g. the probability of up to 2.3% losses within 3 years. Once
we have this quantity, we can derive the corresponding density function.
Knowing this quantity, we can calculate the expected loss of a specific
tranche. This will result in a kind of call option payoff on the loss
probability.
2) From l PF (t , T | M ) , we can derive the payoff of a specific tranche conditional
on the risk factor M . By weighting each payoff with the probability density
of M , we calculate the unconditional expected loss.
Both approaches have their pros and cons. We think that the second one is more
comprehensive. Thus, we elaborate on the second approach in more detail and
just sketch the first one briefly at the end of this section.
The unconditional expected loss of a specific tranche can be calculated via the
following equation:
∞
ELba (t ,Ti , ρ , s, R ) = ∫ Payoff [a, b, l PF (t , T | M )] ⋅ ϕ (M )dM (10)
−∞
Let’s try to explain equation (10). The integral is simply used to remove the
conditional dependency on the state variable M . You can view it as simply
calculating the expected value of the payoff function. ϕ (M ) is the distribution
density function of M , in this case the standard normal distribution function.
This completes the gaussian copula approach, since all stochastic quantities are
assumed to be normally distributed.
Through the payoff function and the upper b and lower a tranche bounds, we
obtain the expected loss of the specific tranche. The payoff can be calculated
using the following equation:


October 5, 2004
Payoff [a, b, l PF (t , T | M )] =
1
b−a
[
(lPF (t , T | M ) − a,0)+ − (lPF (t , T | M ) − b,0)+ ] (11)
This is the payoff profile of a long-call option with strike a and a short-call
option with strike b . Note that the payoff has to be related to the thickness of
the tranche, since the two call options in equation (11) just give the payoff with
respect to the whole portfolio. But we need the payoff with respect to the
tranche thickness. As an example, a portfolio loss of 3% will exhaust the equity
tranche (0% - 3%) completely. Hence, the payoff function for the tranche has to
reach 100% in case of 3% loss in the portfolio.
To obtain some visual insight, we plot the conditional loss fraction l PF (t , T | M ) of
the total index and of some tranches as a function of the level of the risk-factor
M . In the left chart below, we show the loss functions (payoffs) as a fraction of
the total index. As an example, the equity tranche accounts for the first 3% of
portfolio losses and has in this context a thickness of 3%. The relation of this
quantity to the payoff function (see equation 11) is (b-a ) ⋅ Payoff [a, b, l PF (t , T | M )] .
The payoff function of the total index approaches asymptotically the level of
1 − R , which is 50% in the current example. The maximum loss level of a specific
tranche is the thickness b − a of the tranche. It is important to note that these
relative payoffs can be added and subtracted from each other. For example,
adding up the relative payoffs of all tranches results in the loss function of the
index. By subtracting two relative payoff-functions, we can generate a third one.
As an example, the difference between a 0 – 6% equity (relative) payoff and a 0 –
3% equity (relative) payoff will result in a relative payoff of a 3 – 6% tranche.
Note that these calculations refer to relative payoffs, i.e. the quantity
(b-a) ⋅ Payoff [a, b, lPF (t , T | M )] .
EXPECTED LOSS PAYOFF FOR SEVERAL TRANCHES AS A FUNCTION OF THE MARKET FACTOR M
PAYOFF AS FRACTION OF TOTAL INDEX* ABSOLUTE TRANCHE PAYOFF**
50.00% 120.00%
no overlap in tranche payoff
Total index
equity
equity 100.00%
40.00% BBB
BBB
AAA
AAA
80.00% junior
junior
30.00% senior
senior
60.00%
20.00%
40.00%
10.00%
20.00%
0.00% 0.00%
-5 -4 -3 -2 -1 0 1 -4 -3 -2 -1 0 1 2

* E.g. equity tranche has a thickness of 3% relative to the tranche. ** E.g. equity tranche has a thickness of 100%.
In the right chart above, we show the absolute value of the payoff as it can be
directly calculated from equation (11). As we can see by the curves, all payoffs
are adjusted to a maximum loss of 100%. The dashed lines indicate that the
payoff functions do not “overlap”. This means that losses start to erode the
capital of a tranche only if all subordinated tranches are completely exhausted.
Using equation (11) we can solve equation (10) through numerical integration.
Once we can calculate ELba (t , Ti , ρ , s, R ) given all parameters, we have everything


October 5, 2004
we need to price a specific CDO tranche within this model framework.

As indicated above, there is another approach to derive the expected loss
function (the first approach). Let us suppose that F (x, t , T , ρ , s, R ) is the
(unconditional) cumulated probability for a loss fraction x and f (x, t , T , ρ , s, R ) is
the corresponding probability density function. The quantity F (x, t , T , ρ , s, R )
gives the probability that portfolio losses do not exceed the fraction x , while
f (x, t , T , ρ , s, R )dx gives the probability, that portfolio losses will be in the
interval [x, x + dx ] . We can calculate the expected loss of the tranche with
attachment point a and detachment point b in the following way
ELba (t , Ti , ρ , s, R ) = Payoff [a, b, x ⋅ (1 − R )] ⋅ f (x, t , T , ρ , s, R )dx

∫ (12)
0
For more details on the derivation and a closed-form solution for F (x, t , T , ρ , s, R )
and f (x, t , T , ρ , s, R ) in the context of the HLPGC model, please refer to
Schönbucher (2003) p.309ff. Note that this equation is valid for most other
r r r
(
models. We just have to derive the density function f x, t , T , ρ , s , R for this )
model, and solve the integral numerically.
As we introduced above, we need the term structure of expected losses in order
to price a CDO tranche, i.e. we have to know the expected loss up to each
coupon date. Given the formulas derived above, we could easily calculate them
for different Ti . The community discusses another suggestion, which would
simplify the calculation. Derive the expected loss up to the maturity Tn of the
transaction using the formulas above. Then translate this to a spread, and
assume that this spread is constant over time. This results in:
S (t , Tn ) = 1 − ELba (t , Tn , ρ , s, R ) = e − χ ⋅(T −t )
n
(Ti −t ) (13)
S (t , Ti ) = e − χ ⋅(T −t ) = S (t , Tn )(T −t )
i
n
As the second row in equation (13) indicates, this approach essentially comes
down to scaling the expected survival amount up to maturity Tn exponentially by
(Ti − t ) / (Tn − t ) , which obviously differs from 1 − ELba (t , Ti , ρ , s, R ) . From a
mathematical point of view, the time evolution of the expected loss does not
follow the outlined processes, which is mathematically inconsistent.


October 5, 2004
APPENDIX B: PAYOFF FUNCTION FOR BASE CORRELATION

In the concept of “base correlation” we express a compound tranche as the
difference between two equity tranches. In order to do this, we have to express
the expected loss of the compound tranche in terms of the expected losses of the
two equity tranches. This can be done in the following way:
ELba (t , Ti , ρ , s, R ) = ELb0 (t , Ti , ρ , s, R ) − ELa0 (t , Ti , ρ , s, R )

b a
b−a b−a
We show this in the context of the second approach (see Appendix on

mathematical derivation of the HLPGC model).
∫ Payoff [a, b, l
−∞
PF (t ,T | M )] ⋅ ϕ (M )dM
∞
 b 
∫  b − a Payoff [0, b, l (t ,T | M )] − Payoff [0, a, l PF (t , T | M )] ⋅ ϕ (M )dM
a
=
b−a
PF
−∞

∞
∫ Payoff [0, b, l (t ,T | M )] ⋅ ϕ (M )dM

b
=
b−a
PF
−∞
∞
∫ Payoff [0, a, l (t ,T | M )] ⋅ ϕ (M )dM

a
−
b−a
PF
−∞
In the first step we separated the payoff function for the compound tranche into
the payoff functions of the two equity tranches. To understand this step, just
recall that the payoff function is defined a long-call and a short-call payoff.
Payoff [a, b, l PF (t , T | M )] =
1
b−a
[
(lPF (t , T | M ) − a,0)+ − (lPF (t , T | M ) − b,0)+ ]


October 5, 2004
DISCLAIMER
Please note
Key 1: Bayerische Hypo- und Vereinsbank AG and/or a company affiliated with it pursuant to § 15 AktG (German stock-company act) owns at
least 1% of the capital stock of the company.
Key 2: Bayerische Hypo- und Vereinsbank AG and/or a company affiliated with it pursuant to § 15 AktG (German stock-company act)
belonged to a syndicate that in the last five years prior to publication of this analysis has acquired securities of the company.
Key 3: Bayerische Hypo- und Vereinsbank AG acts as stabilizing manager or sponsor, e.g. as designated sponsor of the analyzed securities on
the stock exchange or the open market on the basis of an agreement with the company.
Key 4: Bayerische Hypo- und Vereinsbank AG and/or a company affiliated with it pursuant to § 15 AktG (German stock-company act) hold a
short position of at least 1% of the capital stock of the analyzed company at the end of the month prior to the date on which the
analysis was compiled.
Company - Key: -
Bayerische Hypo- und Vereinsbank AG and companies affiliated with it pursuant to § 15 AktG (German stock-company act) regularly trade
shares of the analyzed company.
Disclaimer:
Our recommendations are based on public information that we consider to be reliable but for which we assume no liability especially with
regard to its completeness and accuracy. We reserve the right to change the views expressed here at any time and without advance notice. The
investment possibilities discussed in this report may not be suitable for certain investors depending on their specific investment target or time
horizon or in the context of their overall financial situation. This report is provided for general information purposes only and cannot be a substitute
for obtaining independent advice. Please contact your bank’s investment advisor. Provision of this information shall not be construed as constituting an
offer to enter into a consulting agreement.
Please note that the provision of investment services may be restricted in certain jurisdictions. You are required to acquaint yourself with any
local laws and restrictions on the usage and the availability of any services described therein. The information is not intended for distribution
to or use by any person or entity in any jurisdiction or country where such distribution would be contrary to local law or regulations.
Notice to UK residents:
This report is intended for clients of Bayerische Hypo- und Vereinsbank AG who are market counterparties or intermediate customers (both as
defined by the Financial Services Authority (“FSA”) and is not intended for use by any other person, in particular, private customers as
defined by the rules of FSA. This report does not constitute a solicitation to buy or an offer to sell any securities. The information in this
publication is based on carefully selected sources believed to be reliable, but we do not make any representation that it is accurate or
complete. Any opinions herein reflect our judgement at this date and are subject to change without notice.
We and/or other members of Bayerische Hypo- und Vereinsbank Group may take a long or short position and buy or sell securities mentioned
in this publication. We and/or members of Bayerische Hypo- und Vereinsbank Group may act as investment bankers and/or commercial
bankers for issuers of securities mentioned, be represented on the board of such issuers and/or engage in “market making” of such securities.
The Bank and its affiliates may also, from time to time, have a consulting relationship with a company being reported upon.
The investments discussed or recommended in this report may be unsuitable for investors depending on their specific investment objectives
and financial position. Private investors should obtain the advice of their banker/broker about the investments concerned prior to making
them.
Bayerische Hypo- und Vereinsbank AG, London branch, is regulated by FSA for the conduct of designated investment business in the UK.
Notice to US residents:
The information contained in this report is intended solely for institutional clients of Bayerische Hypo- und Vereinsbank AG, New York Branch
(“HypoVereinsbank”) and HVB Capital Markets, Inc. (“HVB Capital” and, together with HypoVereinsbank, “HVB”) in the United States, and
may not be used or relied upon by any other person for any purpose. Such information is provided for informational purposes only and does
not constitute a solicitation to buy or an offer to sell any securities under the Securities Act of 1933, as amended, or under any other U.S.
federal or state securities laws, rules or regulations. Investments in securities discussed herein may be unsuitable for investors, depending on
their specific investment objectives, risk tolerance and financial position.
In jurisdictions where HVB is not registered or licensed to trade in securities, commodities or other financial products, any transaction may be
effected only in accordance with applicable laws and legislation, which may vary from jurisdiction to jurisdiction and may require that a
transaction be made in accordance with applicable exemptions from registration or licensing requirements.
All information contained herein is based on carefully selected sources believed to be reliable, but HVB makes no representations as to its
accuracy or completeness. Any opinions contained herein reflect HVB’s judgement as of the original date of publication, without regard to
the date on which you may receive such information, and are subject to change without notice.
HVB may have issued other reports that are inconsistent with, and reach different conclusions from, the information presented in this report.
Those reports reflect the different assumptions, views and analytical methods of the analysts who prepared them. Past performance should
not be taken as an indication or guarantee of further performance, and no representation or warranty, express or implied, is made regarding
future performance.
HVB and any HVB affiliate may, with respect to any securities discussed herein: (a) take a long or short position and buy or sell such securities;
(b) act as investment and/or commercial bankers for issuers of such securities; (c) engage in market-making for such securities; (d) serve on the
board of any issuer of such securities; and (e) act as a paid consultant or adviser to any issuer.
The information contained in this report may include forward-looking statements within the meaning of U.S. federal securities laws that are
subject to risks and uncertainties. Factors that could cause a company’s actual results and financial condition to differ from its expectations
include, without limitation: political uncertainty, changes in economic conditions that adversely affect the level of demand for the company’s
products or services, changes in foreign exchange markets, changes in international and domestic financial markets, competitive
environments and other factors relating to the foregoing.
All forward-looking statements contained in this report are qualified in their entirety by this cautionary statement.


October 5, 2004
CONTACTS
Global Head of Research
Thorsten Weinelt, CFA
Managing Director
+49 89 378-15110
thorsten.weinelt@hvb.de
FX/FI & FX/FI Derivatives Strategy High Grade Research*

Michael Rottmann, Head Luis Maglanoc, CFA, Head
+49 89 378-15121 Financials, Sub-Sovereigns
+49 89 378-12708
Kornelius Purps, FI-Strategy
+49 89 378-12753 Elena Guglielmin
Financials
Herbert Sellier, Technical Analysis
+49 89 378-16296
+49 89 378-18024
Stephan Haber
Herbert Stocker, Technical Analysis
Telecoms, Media, Technology
+49 89 378-14305
+49 89 378-15192
Armin Mekelburg, FX-Analysis
Franz Rudolf
+49 89 378-14307
Utilities, Energy
Dr. Stefan Kolek, EEMEA Strategy +49 89 378-12449
+49 89 378-12495
Dr. Sven Kreitmair
Frauke David Automobiles, Industrial G&S, Aerospace
+49 89 378-13247 +49 89 378-13246
Carmen Hummel
Non-Cyclical G&S, Food & Beverage, Cyclical G&S,
Credit & Credit Derivatives Strategy
Retail
Dr. Jochen Felsenheimer, Head +49 89 378-12252
+49 89 378-18188
Dr. Philip Gisdakis
High Yield* & EEMEA Credit Research
Quantitative Credit Strategy
+49 89 378-13228 Dr. Felix Fischer, CFA, Head
General Industries, Basic Resources, Construction,
Michael Zaiser
Tobacco
Credit Strategy
+49 89 378-15449
+49 89 378-13229
Jochen Schlachter
Chemicals, Healthcare
Covered Bond & Agency Research +49 89 378-13212
Fritz Engelhard, Head Jana Arndt
+49 89 378-18133 +49 89 378-13211
Florian Hillenbrand Dusan Meszaros, EEMEA Credit
+49 89 378-12960 +43 505 058-2350
dusan.meszaros@BA-CA.com
Securitization Research
Helge Münkel
+49 89 378-11294 * Crossover credits are covered by the
respective high-grade sector analyst
Publication Address
Bayerische Hypo- und Vereinsbank AG
Bloomberg
HVB Corporates & Markets
Global Markets Research HVBR
Arabellastrasse 12
D-81925 Munich
Internet
Telephone +49 89 378-12759
Facsimile +49 89 378-16237 www.hvb.de/valuepilot

HVB Group DJ ITRAXX Credit at Its Best

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

HVB Group DJ ITRAXX Credit at Its Best

Uploaded by

Copyright:

Available Formats

Global Markets Research

European Credit Strategy

Credit Derivatives Special

DJ ITRAXX: CREDIT AT ITS BEST!

HVB CDS/DJ iTraxx Trading

HVB DJ iTraxx Marketing -40%

Source: HVB Global Markets Research

n Trading the DJ iTraxx: The DJ iTraxx universe offers the opportunity to

© HVB Corporates & Markets, Global Markets Research.

Credit Derivatives Special - DJ iTraxx: Credit at its best!

THE ITRAXX FAMILY: A WIDE RANGE OF INSTRUMENTS AVAILABLE*

DJ iTraxx Europe Sub-Indices

Source: HVB Global Markets Research

© HVB Corporates & Markets, Global Markets Research. 2

Credit Derivatives Special - DJ iTraxx: Credit at its best!

© HVB Corporates & Markets, Global Markets Research. 3

Credit Derivatives Special - DJ iTraxx: Credit at its best!

DJ ITRAXX EUROPE HIVOL INDEX

DJ ITRAXX EUROPE CORPORATE

© HVB Corporates & Markets, Global Markets Research. 4

Credit Derivatives Special - DJ iTraxx: Credit at its best!

DJ ITRAXX EUROPE CROSSOVER

© HVB Corporates & Markets, Global Markets Research. 5

Credit Derivatives Special - DJ iTraxx: Credit at its best!

STRUCTURE OF STANDARDIZED ITRAXX TRANCHES

DJ iTraxx Europe 5Y - Reference Pool

Equity* BBB* AAA*

Junior Super Junior Super

Investment grade* (3-100% / 97% / 3%)

*in brackets: (tranches / thickness / subordination)

© HVB Corporates & Markets, Global Markets Research. 6

Credit Derivatives Special - DJ iTraxx: Credit at its best!

DJ ITRAXX EUROPE FIRST TO DEFAULT BASKETS

© HVB Corporates & Markets, Global Markets Research. 7

Credit Derivatives Special - DJ iTraxx: Credit at its best!

© HVB Corporates & Markets, Global Markets Research. 8

Credit Derivatives Special - DJ iTraxx: Credit at its best!

TRADING THE ITRAXX

FUNDED VERSUS UNFUNDED AND THE P&L IMPACT

DJ ITRAXX SWAP FAMILY: SERIES 1 AND 2 COUPONS

DJ iTraxx index Maturity Coupon – Series 1 Coupon – Series 2

© HVB Corporates & Markets, Global Markets Research. 9

Credit Derivatives Special - DJ iTraxx: Credit at its best!

with a short position today. To keep the premium constant, an investor

© HVB Corporates & Markets, Global Markets Research. 10

Credit Derivatives Special - DJ iTraxx: Credit at its best!

PRICING AND QUOTATION OF DJ ITRAXX SWAPS

ISSUE NO. 1 - THE “DISPERSION BIAS”

no default after A defaults after B defaults

© HVB Corporates & Markets, Global Markets Research. 11

Credit Derivatives Special - DJ iTraxx: Credit at its best!

A SIMPLE PRICING MODEL FOR BASKET CDS

© HVB Corporates & Markets, Global Markets Research. 12

Credit Derivatives Special - DJ iTraxx: Credit at its best!

with R X = (expected) recovery rate for name X

At initiation of a CDS contract in t = 0, the spread fixing is governed by the

© HVB Corporates & Markets, Global Markets Research. 13

Credit Derivatives Special - DJ iTraxx: Credit at its best!

WEIGHTED VERSUS UNWEIGHTED AVERAGE

with q X (ti ) = probability for name X to survive time ti without default

© HVB Corporates & Markets, Global Markets Research. 14

Credit Derivatives Special - DJ iTraxx: Credit at its best!

THE FAIR SPREAD OF A CDS INDEX