Q U A N T I T A T I V E C R E D I T R E S E A R C H

Quantitative Credit Research

Quarterly

INTRODUCTION
RESEARCH
Over the last several years, there has been a significant increase in the importance
of the global credit markets to institutional investors. This has been driven by a
number of different factors, including the lower supply of government debt and
the desire to enhance long-term risk-adjusted returns. All of this growth has coin-
cided with the rapid expansion of the credit derivatives market and has heightened
the demand for a better understanding of what drives credit spreads and how these
dynamics then affect their nearest derivative proxy: credit default swaps.

This growth of the credit derivatives market has also created a demand for a
better understanding of the new products that have emerged. In particular, de-
fault correlation products such as default baskets are gradually being seen as a
new asset class to sit alongside single-name credit products.
October 18, 2001

Arthur M. Berd The first article in this publication examines the drivers of credit spreads. The
[email protected] authors break these out into a number of fundamental factors, including default
risk, re-investment risk, macroeconomic effects, risk-aversion, and liquidity, as
Mark W. Howard well as technical factors such as off-equilibrium supply-demand pressures. This
[email protected] analysis leads to the conclusion that the current very high levels of credit
spreads are driven by historically high levels of risk premia rather than
Robert McAdie
44-20-7260-3036 historically high levels of risks. The authors envision substantial spread tight-
[email protected] ening once the overhang of the war, recession, and risk aversion is removed.

Marco Naldi The second article attempts to describe how investors may begin to take advan-
[email protected] tage of the opportunities presented by the credit derivatives market to trade the
Dominic O'Kane spread between the cash and credit default swap market. The authors set out the
44-20-7260-2628 various factors that cause the cash and credit default swap market to dislocate as
[email protected] well as the implications of the various ways one can implement a default swap
basis trade. Using a model, they examine the theoretical relationship between
Lutz Schloegl the carry of the basis trade and any gain or loss in the event of default. A number
44-20-7601-0011, ext. 5016
of scenarios are presented that demonstrate the dynamics of the default swap
[email protected]
basis and how investors can exploit this as a relative value opportunity.

For many investors, it is unclear what, if any, advantages basket default swaps
present over single-name credit derivatives. In response, the authors of the third
Quantitative Credit

article introduce two metrics to analyze baskets. The first is the excess spread,
defined as the difference between the spread paid by a default basket and the
breakeven spread computed from historical default studies. The second is the
spread coverage ratio, which is simply the ratio of these two quantities rather
than the difference. It represents the number of times the spread is compensat-
ing the issuer for the pure default risk of holding the product. The authors find
that first-to-default baskets are an ideal way to leverage excess spread. For the
spread paid, they present a lower expected loss for single-name assets paying a
similar spread. For investors who wish to maximise the spread coverage ratio,
second-to-default are better than single-name credit derivatives, especially for
pools of credits with a low default correlation.

Within the world of quantitative credit, one of the hottest problems of the mo-
ment is determining the best way to model correlated defaults. Such a model is
essential for valuing and risk-managing the family of portfolio credit products
that includes first-to-default baskets and both cash flow and synthetic CDOs.
This question is examined in detail by the authors of the last article. Many of the
models currently used in the market adopt a multi-variate normal distribution for
asset returns in order to generate correlated defaults. In their article, the authors
demonstrate that such an assumption significantly underestimates the likelihood
of extreme events. They then propose an alternative model that allows for ex-
treme movements and discuss how its parameters can be calibrated to agree with
empirical data. The effect on the pricing of first and second-to-default baskets is
presented and discussed.

2 Lehman Brothers, October 18, 2001

 CONTENTS

Spreading the Praise and the Blame .................................................................... 4

Trading the Default Swap Basis ......................................................................... 13

Leveraging Spread Premia with Default Baskets .............................................. 22

Extreme Events and the Valuation of Default Baskets ..................................... 34

Lehman Brothers, October 18, 2001 3

Quantitative Credit

Arthur M. Berd SPREADING THE PRAISE AND THE BLAME

[email protected]
We analyze the fundamental and technical catalysts behind credit spread move-
Mark W. Howard
[email protected] ments and discuss the outlook following the terrorist attacks in the US.

Introduction
The global credit markets have become increasingly important in the asset allo-
cation framework for institutional investors over the last five years. The U.S. has
the best-developed capital markets for investment-grade and high-yield corpo-
rate bonds, but both Europe and Asia are developing rapidly. Figure 1 shows the
composition of the Lehman Brothers U.S. Aggregate Index by amount outstand-
ing in major asset classes. Corporate bonds have significantly increased in
importance since 1995, mostly at the expense of U.S. Treasury bonds, which
have declined recently while the government retired its debt.

Investors outside the U.S. are rapidly expanding their usage of credit product
in order to enhance long-term risk-adjusted returns. Additionally, origina-
tion in the convertible securities market has grown substantially over the last
18 months. The market for structured credit product has doubled in each of
the last three years and now exceeds $1 trillion. And with global interest
rates steadily ratcheting lower and banks tightening their lending standards,
the importance of the public credit markets has never been greater for the
world’s major corporations and sovereigns.

Consequently, a thoughtful interpretation of credit spread behavior (as opposed

to swap spreads, which are no longer a consistent credit proxy) has significant
meaning for debt investors, borrowers and equity market participants alike.

We would like to thank our colleagues Ivan Gruhl and Stephen Mandl for their contributions to this
article.

Figure 1. Lehman Brothers U.S. Aggregate Index Composition

by Amount Outstanding

7,000,000 Lehman Aggregate Index Composition

6,000,000
Amt. Outstanding ($MM)

23%
5,000,000
21%
4,000,000 17%
17%
3,000,000

2,000,000

1,000,000

0
Aug-1992 Aug-1995 Aug-1998 Aug-2001

US Treasury US Agency MBS IG Corporate Other

4 Lehman Brothers, October 18, 2001

 What Drives Credit Spreads?
Our goal in this article is to analyze some of the relationships that govern the
behavior of credit spreads, concentrating on the U.S. investment-grade corpo-
rate market. There is a similarity between current market conditions – following
the September 11 terrorist attacks in the US – and those during the Gulf War. As
a result, we have extended our analysis to the decade since the last recession.
Since 1990, we have seen several important cycles in the behavior of the credit
spreads.

Like any other asset class, credit product offers investors potential incremental re-
turn for taking on particular types of risk. The single parameter that describes the
return side of this relationship is the credit spread and its behavior (tightening or
widening).

The extra yield offered by credit product compensates the investor for a variety of
risks. What are the particular risks that are associated with credit instruments?

• Credit risk: The risk that the investor will not receive the full amount of
the debt repayment as promised due to default of the issuing company or
non-corporate entity.

• Reinvestment risk: Unless the investors intend to hold the bonds until fi-
nal maturity and unless they are insensitive to the mark-to-market of their
holdings, such an investor bears the risk of spread level changes, even if the
asset specific default risk is constant.

• Technical, or supply/demand risk: The supply/demand pressures can be

significant in the short and intermediate terms, as they distort the risk and
return patterns associated with normal equilibrium state of the market (es-
pecially during crisis periods).

• Macroeconomic risk, liquidity risk, and risk aversion: As the economy

cycles through expansions and recessions and weathers cataclysmic events
such as terrorist attacks or financial market disruptions, the credit markets
follow in a closely related manner. The importance of macro risk is in chang-
ing levels of risk premia that investors demand for taking on asset-specific
risks. The liquidity premium, while not identical, is closely related to risk
aversion and reflects the added premium for holding securities other than cash
or highly liquid treasuries.

The Credit Risk Factor

In terms of default risk, the U.S. credit markets in general, and the high-yield
market in particular (which bears the vast majority of default risk inside of a one-
year time horizon), have gone through a full cycle during the last 10 years. Figure
2 shows the rolling 12-month default experience in the U.S. high yield markets.
Default rates have more than doubled since the mid 1990s.

Lehman Brothers, October 18, 2001 5

Quantitative Credit

Figure 2. U.S. Corporate Spreads versus

Default Rates and Spread Volatility

250 35%

30%

default rates, spread volatility

200
25%

spread (b.p.)
150 20%

100 15%

10%
50
5%

0 0%

Aug-90

Sep-91

Sep-92

Sep-93

Sep-94

Sep-95

Sep-96

Sep-97

Sep-98

Sep-99

Sep-00

Sep-01
Investment Grade OAS Spread Volatility
Default Rates

Moreover, recovery values have deteriorated substantially, leading to a markedly

higher default loss risk, which almost tripled during the same period. Because of
the “new economy” nature of many recent bankruptcies, there is much less that
can be salvaged by distressed investors.

The realized default rates are, by construction, lagging variables. We have shown
them on a 6-month forward-looking basis; i.e., each data point refers to the real-
ized numbers over the period covering six months prior to six months after. Figure
2 shows a remarkable accuracy with which the credit markets discount the future
behavior of the credit risk factor—note the synchronicity of the pattern of spreads
with future realization of default losses.

The Reinvestment (Volatility) Risk Factor

On the same Figure 2, we show the behavior of corporate spreads versus spread
volatility.1 We can clearly see four major volatility cycles: the first in the early
1990s during the U.S. recession and Gulf War, the second in 1997 after the Asian
crisis, the third following the Russian crisis and hedge fund debacle in 1998, and
the fourth unfolding during early 2000 with the burst of the technology bubble.
Interestingly, realized spread volatility has not reached 1998 levels yet, helping
to explain the excellent performance of credit markets in the first half of 2001.

In all probability, at present we are witnessing the beginning of another volatility

cycle, with the U.S. economy slipping into recession and the prospect of a pro-
longed military conflict compounding domestic problems. This has contributed
to recent spread widening and will put a floor on any recovery until the reinvest-
ment risk subsides to acceptable levels.

1 We measure spread volatility by an exponentially weighted moving estimate with a 6-month half-life.

6 Lehman Brothers, October 18, 2001

 The Supply-Demand Factor
Given its size and liquidity, the Lehman Brothers Credit Index is a good repre-
sentation of the credit marketplace. Figure 3 shows investment-grade corporate
bond supply and demand due to international investor buying, as well as reinvest-
ment of coupon and redemption flows from the Credit Index. We also show
secondary trading volumes, as measured by Lehman Brothers' High-Grade cash
corporate bond trading desk. The net demand for credit product, together with
trading volumes and spread volatility, can be combined into a market impact vari-
able, as explained in the Appendix.

A potentially large component of the demand for credit products is due to asset
reallocation in balanced portfolios. However, we believe that this component
can be ignored in our simplified framework because it is characterized by a sub-
stantially longer typical time scale. Large institutional investors, such as pension
funds, normally make asset allocation decisions annually. Others, such as bal-
anced mutual funds may adjust their allocations semi-annually or quarterly. Of
course, large market dislocations like the 1998 credit crunch or the current situ-
ation do cause a revision in allocation decisions. Thus, we contend that during
most of the time (except major crises), the effect of asset reallocations on
monthly spread changes is relatively small.

Figure 3. U.S. Credit Index Supply, Estimated Demand and Lehman Brothers’
Trading Volumes
120 18
100 15
supply and demand ($B/mo)

trading volume ($B/mo)

80 12
60 9
40 6
20 3
0 0
-20 -3
-40 -6
Nov-99

Nov-00
Jan-99

Jul-99

Jan-00

Jul-00

Jan-01

Jul-01
Mar-99

May-99

Sep-99

Mar-00

May-00

Sep-00

Mar-01

May-01

Foreign Demand Reinvestment Demand

Corp Supply Trading Volume

Lehman Brothers, October 18, 2001 7

Quantitative Credit

Figure 4. Market Impact of Net Demand

IG Corprate Spread Changes vs. Net Demand Market Impact
0.4 0.4
0.3 0.3

0.2 0.2

spread change (%)

Market impact (%)
0.1 0.1

0.0 0.0
-0.1 -0.1

-0.2 -0.2

-0.3 -0.3
-0.4 -0.4

Jan-99

Mar-99

May-99

Jul-99

Sep-99

Nov-99

Jan-00

Mar-00

May-00

Jul-00

Sep-00

Nov-00

Jan-01

Mar-01

May-01

Jul-01
Mkt Impact Change in OAS

Figure 4 shows the projected impact of the net demand for credit products for
the U.S. investment-grade corporate sector versus the month-over-month changes
in the average OAS of the index. It is important to note that a net positive supply
does not always lead to observable widening of credit spreads because it is the
issuer that bears the cost of the market impact by pricing bonds at wider spreads.
This phenomenon leads to new issue rallies that are not accounted for in the
Aggregate Index because the bonds are included in the index only from the month
following issuance. On the other hand, in the case of unbalanced positive de-
mand, there are no other counterparties except investors, and, therefore, the
market impact is not shielded.

Figure 5. Lehman Corporate Index OAS vs. UST Slope and Liquidity Premium

300

250

200
spread (b.p.)

150

100

50

-50
Aug-89

Sep-90

Sep-91

Sep-92

Sep-93

Sep-94

Sep-95

Sep-96

Sep-97

Sep-98

Sep-99

Sep-00

Sep-01

IG Corporate OAS UST 2-10 Spread Liquidity Premium

8 Lehman Brothers, October 18, 2001

 The Macro Factor
The U.S. Treasury curve slope has been a good proxy for the macro factor during
normally functioning markets, with the inverted curve in early 2000 signaling a
future recession risk.1 As the Fed started reducing interest rates, the curve re-
verted to its normal upward sloping shape, reflecting the market’s expectations
of an economic recovery. As Figure 5 shows, investment-grade credit spreads
usually widen when the Treasury curve is inverted and tighten when it subsequently
steepens.

We also show the Lehman Brothers’ LIBOR Credit Liquidity Premium on the
same chart. It has been on the increase since the 1998 Russian crisis and cur-
rently stands at 50 bp, more than twice its value in 1997. These extra 25 basis
points could potentially contribute to spread rallies when the more immediate
political and economic risks subside.

Compared with the previous recession and war period of 1991, we observe a
promising similarity in the behavior of the U.S. Treasury curve slope. We can
also see that the positive slope of the curve is still almost 100 bp below the highs
reached in late 1992, when the economic recovery was complete and the new
expansion had started. It is likely to take both additional Fed rate cuts and signs
of expected recovery (reflected in higher 10-year yields) to get to those levels.

It seems plausible that not only the value of the US Treasury slope, but also the
amount of time during which this value is negative or very low may be important.
Indeed, a hypothetical “deep recession” scenario when the curve stays inverted
for a long time might develop if the Fed easing were not aggressive enough in
pushing short rates below long rates that are depressed by prospects of the slow-
ing economy. On the other hand, a hypothetical “soft landing” scenario could
develop when curve inversion happens as a short-term phenomenon due to very
active Fed tightening followed by a quick subsequent easing without affecting
long term interest rates.

To quantify these possibilities we introduce the integral macro factor. Let us

consider the long term average of the US Treasury 2-10 spread as its indicative
equilibrium value. In Figure 6, we show the difference between the current value
of the slope and its long-term average, shading it in dark gray when the slope is
less than the average and light gray when it is greater than the average. The inte-
gral macro factor is defined as the cumulative shaded area between the spot value
of the slope and its long-term average (with proper sign). This definition cap-
tures the intuitive notion outlined above that the longer the slope stays inverted
(or not steep enough), the greater the Fed must lower rates (as manifested by a
large positive slope) to reverse the impact of a slowing economy.

1 We proxy the Treasury curve slope by the 2s-10s, i.e. the difference between the yields of the 10-year
and 2-year Treasury bonds.

Lehman Brothers, October 18, 2001 9

Quantitative Credit

Figure 6. Defining the Integral Macro Factor

3.0 3.0

2.5 2.5

2.0 2.0

Integral Macro Factor (%)

UST 2-10 Spread (%)
1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0

-0.5 -0.5

-1.0 -1.0

-1.5 -1.5
Jul-87

Jul-88

Jul-89

Jul-90

Jul-91

Aug-92

Aug-93

Aug-94

Aug-95

Aug-96

Sep-97

Sep-98

Sep-99

Sep-00

Oct-01
UST 2-10 Spread
Long-term average of 2-10 Spread
Integral Macro Factor (Cumulative Area)

The integral macro factor becomes negative during the recession of 1991 when
the slope dips below the average, then recovers to the positive territory when the
curve steepens in 1992, and then gradually decreases in value as the US Treasury
curve increasingly flattens in recent years, finally resulting in negative slope and
negative integral macro factor in the second half of 2000. Interestingly enough,
the depth of the recent dip in the integral macro factor is considerably less than
what happened in the prior recession. While the US economy has certainly missed
the “soft landing” scenario, Figure 6 holds out hope that the “deep recession”
scenario may not be very probable either. As Lehman Brothers’ economists pre-
dict, we are likely to see a relatively shallow recession. Of course, the
uncertainties still abound after the terrorist attacks of September 11, and all these
indications must be taken with a grain of salt.

What to Expect in the Next Year?

As we have discussed above, credit spread movements have become more difficult
to predict over the last few years as the web of inter-connecting influences has
grown more complex. However, we believe that a thorough understanding of the
various fundamental and technical determinants of spread activity will allow mar-
ket participants to navigate today’s challenging markets.

One important question on credit market participants’ minds is whether the cur-
rent levels of spreads are the new norm or if a return to the levels of pre-1998 is
feasible. A comparison of the current situation with the 1990-91 era has merit
due to the similarity of economic and political conditions. However, the un-
stable nature of the current capital markets renders our predictive vision limited.

10 Lehman Brothers, October 18, 2001

 It certainly appears that current spread levels are incongruously high compared
with those of 1990-91. Indeed, default rates have not yet reached prior levels and
may remain under those levels even if the recession materializes. Volatility risk
is noticeably lower and is unlikely to reach the extremes of 1998 credit crunch.
The supply/demand equation is quickly moving into net positive supply over the
short term, with many companies boosting their financial leverage. Finally, the
Fed is accommodating aggressively, and we will certainly see a much stronger
fiscal and monetary stimulus than was previously expected.

Thus, much of the current level of spreads is due not as much to the historically
high levels of risk, but to historically high levels of risk aversion and risk premia.
Therefore, when the overhang of the recession and the international hostilities is
removed, one should expect credit spreads to narrow substantially. The critical
issue for investors, as always, is to recognize and benefit from the beginning of
this process.

Appendix A. Computing Supply/Demand Impact on Spreads

In this Appendix, we outline a simple toy model that estimates the market impact
in terms of spread widening or tightening induced by the supply/demand pres-
sures. There are many dimensions to this problem. In its full complexity, the
determination of the market impact is among the toughest challenges in eco-
nomics and finance, because almost all standard assumptions, such as equilibrium,
market efficiency, frictionless trading, symmetrical information among the market
participants, etc. need to be explicitly abandoned.

In the academic literature, many have focused on modeling the market micro-
structure in either game-theoretic or asymmetric information markets setting.
Among the industry practitioners, a more popular choice is the models, which
focus on simulations of market clearing mechanisms and liquidation-equivalent
prices of instruments. This approach has been particularly well studied in case of
equity markets, where the best know example is Barra Market Impact model.

We, on the other hand, will approach the problem in a simplified manner. Our
definition of the market impact will be limited to the determination of the fair
spread premium or discount, which a hypothetical dealer would impose in order
to “work the order” representing the aggregate net supply or demand. We will
assume that such dealer has a good sense of the average trading volumes in the
market, and does not wish to take any particular view on the direction of the
market. Moreover, we will assume that there is no tradeable information content
in the off-equilibrium net demand. This is clearly an oversimplification, but we
believe that when considering a model for the market aggregate net demand all
the client-specific information can be ignored.

Our hypothetical dealer’s main guiding rule is to have appropriate risk budget to
cover his potential shortfall while making the trades. Thus, our model must de-
pend on the net demand for credit product, the trading volume and spread volatility
as the primary observable variables. Assume that in a given month the net demand

Lehman Brothers, October 18, 2001 11

Quantitative Credit

for credit assets is ND , the level of the market activity is proportional to the
secondary market trading volume V , and that the spread volatility is σ . Posi-
tive net demand will cause spread tightening, while negative net demand (i.e.
positive net supply) will presumably cause spread widening. Everything else be-
ing equal, what is the fair spread differential that an investor should give up for
entering into such market?
The risk due to off-equilibrium net demand is that by the time it takes to
complete the trading, spreads will have moved and the average spread of
purchased bonds will be tighter than when trading began. It will take
approximately T = ND V time to digest the demand.
The fair cost of possibility that the volatile security will have a value greater
(lower) than its current level after such time is proportional to the at-the-
money call (put) option price with maturity equal to trading horizon. The
corresponding option premium can be written as follows:

Premium = Price underlying ⋅ Price option (Spot = 1, Strike = 1, σ price , T )

We then must interpret the option premium in the equation above as the
potential addition (reduction) to the price of the bond at the inception of the
hypothetical trade, which is necessary to cover the dealer’s risk budget. For
high grade fixed income bonds, the price is near par. We denote the fair
premium (discount) in spread terms as ∆s . It is related to price premium via
usual duration relationship:

Premium
= − D ⋅ ∆s
Price underlying
Similarly, the price volatility can be re-expressed in terms of the spread
volatility as follows:

σ price = D ⋅ σ spread
For both at-the-money puts and calls, the dependence on trading horizon is
well approximated by the option theta, which leads to the following simple
market impact formula for short horizons:

⋅ Θ ATM (σ price , T ) ⋅ T
ND
∆s ( ND,V , σ ) ∝ − sign ( ND ) ⋅ ≈ −sign (ND ) ⋅ σ spread
1
D T=
ND V
V

The duration dependence cancels out. The fair cost is proportional to the
cumulative risk during the trading horizon σ T .

12 Lehman Brothers, October 18, 2001

 Robert McAdie TRADING THE DEFAULT SWAP BASIS
Credit Strategy
[email protected] We analyse the new relative value opportunities presented by the credit deriva-
44-20-7260-3036 tives market to trade the spread between the default swap and cash market.

Dominic O’Kane The advent and significant growth of the credit derivatives market has presented
Quantitative Research investors with new ways to trade credit risk. In addition it has created new relative
[email protected] value opportunities. One such opportunity is the trading of the default swap basis
44-20-7260-2628 in which investors may take a relative value view on the spread between a bond
issued by some entity, and the spread demanded by a default swap contract linked
to that same entity. While there is a theoretical relationship between these two
spreads, there are a number of factors that may cause this relationship to break

Figure 1. The Theoretical Risk-Free Trade

Before we explain why default swaps and cash can trade at different spread lev-
els, we must first establish why they are related. To do this, we describe what we
call the theoretical risk-free trade. This is a strategy involving a cash bond and a
default swap which subject to certain assumptions can be shown to be risk-free.
We consider a long basis trade, consisting of:

1) Purchase of a defaultable floating rate note worth par which pays a

coupon of Libor plus a fixed spread F. We assume that the asset is
funded at Libor flat.
2) Purchase of protection on this asset in the default swap market to its
maturity at a default swap spread D.

This strategy is shown in Figure 1. In the event of default the investor delivers the
defaulted asset to the protection seller in return for par. This par amount is then
used to pay off the funding leg. The net strategy is therefore credit risk free, as the
investor has no exposure to the default of the asset. The investor earns an annual
spread of +(F-D) over Libor for assuming no credit risk. The theoretical arbitrage-
free relationship requires that D=F.

Static hedge for a protection buyer showing a) The payments before and b) in the
event of default

Protection Seller Protection Seller

Default
Swap Defaulted 100
spread D Asset
Pay
100

Hedged Investor Defaulted Hedged Investor

Asset (protection buyer) Asset (protection buyer)

LIBOR
+F
Repay
Borrows 100
100 LIBOR

Funding Funding

(a) Before default or maturity (b) At Default

Lehman Brothers, October 18, 2001 13

Quantitative Credit

down. In certain cases this can present clear investment opportunities to investors.

In an earlier publication1 we set out in detail the reasons why cash and default
swap spreads may differ. In what follows, our aim is to provide market partici-
pants with a framework for understanding how to analyse and implement basis
trading strategies. Using a model we establish the theoretical relationship be-
tween cash and default swaps. We then discuss how one might transform a view
about the default swap basis into a specific trading strategy, discussing how this
theoretical relationship can break down in practice. Before doing so we define a
long basis trade as one in which we buy the asset and buy protection. A short
basis trade is one in which we sell the asset and sell protection. The theoretical
relationship between a par floater and a default swap is shown in Figure 1.

Implementing a long basis trade

The standard way for a credit investor go long the basis is to purchase the bond
on asset swap and then to buy protection. In a standard par asset swap, the inves-
tor pays par to buy a package consisting of the bond plus an interest rate swap to
exchange the fixed coupon on the bond for payments of Libor plus a fixed spread
known as the asset swap spread.

How precisely the credit risk of this position should be hedged depends on a num-
ber of factors, most important of which is the initial full price (clean price plus
accrued interest) of the bond. As a default swap is a par product it can only hedge
the difference between par and the recovery value of the asset. This is not a prob-
lem if the asset is initially priced at par. However, if the full price of the asset is
trading away from par, a default swap does not exactly hedge the initial investment.

To see this in more detail, consider the case of an investor who purchases a fixed
rate bond on asset swap, and buys protection with a default swap on the full face
value of the bond to maturity. We call this a face value hedge. We assume that
the cost of funding is Libor flat.

1 Trading the Basis: Cash vs Default Swaps Dominic O’Kane, Robert McAdie, Lehman Publication,
Jue 2001.

Figure 2. Bond on asset swap plus face value hedge showing flows for different events

Event Asset Asset Swap Funding Default Swap Net

Swap

Initiation -Full Price -(100%-Full Price) +100% 0 0

Carry +Coupon -Coupon+(Libor+Asset Swap Spread) -Libor -Default Swap Spread +(Asset Swap Spread
-Default Swap Spread)

Immediate Default +Recovery +(100%-Full Price) -100% (100%-Recovery) +(100%-Full Price)

Default at time t +Recovery Swap MTM(t) -100% (100%-Recovery) Swap MTM(t)

Maturity +100% 0 -100% 0 0

14 Lehman Brothers, October 18, 2001

 As shown in Figure 2, the annual carry for the investor in this trade is given by the
difference between the asset swap spread and the default swap spread.

In the event that the asset defaults, the investor is left with a defaulted asset and
an off-market swap, which must be unwound. Initially, the value of the interest
rate swap equals 100%-Full Price, but this changes over time as interest rates
change and as the swap rolls down the Libor curve. If the asset is a premium
asset, the mark-to-market will initially be negative. If the asset is a discount as-
set then this swap will initially be worth a positive amount and any default will
result in a positive payment from the swap. As maturity is approached, the value
of the swap pulls to zero so that the net position equals par in the event of default.

Buying the asset on asset swap has the benefit that it removes most of the inter-
est rate exposure that would be incurred by buying the fixed rate bond. It removes
much of the price variation of the bond which result from movements in the
Libor curve and the fixed coupon payments. This means that, on a mark-to-mar-
ket basis the investor's default exposure changes less for an asset swap than for a
fixed coupon bond. For example, the default exposure profile for a fixed coupon
discount bond hedged on the full face value with a default swap is shown in Fig-
ure 3. To hedge this exposure, the investor would have to dynamically buy or sell
default protection throughout the life of the trade. Transaction costs may make
this impractical, especially given the small sizes involved. Instead, the investor
may choose to hedge a slightly higher notional at the cost of losing some carry.
For this reason, asset swaps are preferable.

The exception to this is bonds trading at a deep discount. These tend to trade on
a price basis, where the recovery value of the asset becomes more critical and
where the bond price becomes less sensitive to spread or interest rate move-
ments. In this case investors may prefer to buy the bond outright and buy protection
without entering into an asset swap.

Figure 3. The change in gain in the event of default for a face-value hedged
5% coupon bond as the coupon payments are made and as the
bond rolls down the yield curve
6.0%
Gain in the Event of Default (% of

4.0%

2.0%
Facevalue)

0.0%
0.00 1.00 2.00 3.00 4.00
-2.0%

-4.0%

-6.0%
Time (years)

Lehman Brothers, October 18, 2001 15

Quantitative Credit

An alternative hedging approach for the investor is to determine the amount of

default protection in order to protect their initial investment. We call this a
market-value hedge. It means buying protection on a notional equal to the mar-
ket hedge ratio times the face value of the cash where
Full Price-Expected Recovery
Market Hedge Ratio =
100%-Expected Recovery

The effect of this approach is to change the carry on the position. The gain on
default is initially zero provided our assumption of the expected recovery rate is
correct. If default does occur and the actual recovery rate is different from the
expected recovery then the gain on default is

 Full Price-Expected Recovery 

(1-Actual Recovery)   -(Full Price-Actual Recovery)
 100%-Expected Recovery 

For a discount bond this means that we lose if the actual recovery is less than
the expected recovery, and gain if the actual recovery is greater than the ex-
pected recovery. For a premium bond we lose if the actual recovery is greater
than the expected recovery, and gain if the actual recovery is less than the ex-
pected recovery. We therefore have a recovery rate risk, which can be in our
favour or against us.

A third type of hedge is to buy a notional of protection equal to the market price
of the bond. This is equivalent to the market-value hedge with an expected recov-
ery rate of zero. We therefore call this a zero recovery market-value hedge.
The cost of protection for a discount bond is between that of a face value hedge
and a market-value hedge. It guarantees that an investor can never lose more than
their initial investment. However, it still leaves the investor underhedged on a
mark-to-market basis as the asset accretes up to par. We view the zero-recov-
ery market hedge as an attractive compromise between the face value and
market-value hedging strategies since it eliminates the downside recov-
ery risk and has greater carry than a face value hedging strategy for a
discount bond.

Using a simple model of default and recovery we can establish a theoreti-

cal relationship between the default swap spread and the asset swap spread.
In Figure 4(a) we have plotted the annualised carry versus the gain in the
event of default for a 5-year asset with a coupon of 6%, on asset swap. This
is done by varying the default probability of the issuer, and is shown for all
three hedging strategies.

The first thing to note is that all three lines pass through the origin. This
corresponds to the case when the asset prices at par in which case all three
hedging strategies are equivalent and the carry is zero. The market value
hedge is a vertical line since, assuming that the actual recovery rate equals
the expected recovery rate, there should be zero gain in the event of an
immediate default. We can see this more clearly in Figure 4(b) where we

16 Lehman Brothers, October 18, 2001

 Figure 4a. Annualised carry versus gain in the event of an immediate
default for three hedging strategies

60 Facevalue Hedge

40 Market Value Hedge

Annual Carry (bps)

20 Zero recovery Market Value
Hedge
0
-10% -5% 0% 5% 10% 15%
-20

-40

-60

-80
Gain in the event of immediate default as a percentage of
par

Figure 4b. Annualised carry versus bond price. The bond chosen is a
5-year 6% coupon bullet bond 60

40

20

Annual Carry (bps)

0
110% 105% 100% 95% 90% 85%
-20
Facevalue Hedge
-40
Market Value Hedge

Zero recovery Market Value -60

-80
Price of the bond as a percentage of par

have plotted the carry versus the bond price. Once again all three lines in-
tersect at a price of par. The carry for the market value hedge increases as
the bond price falls below par, reflecting the fact that while we are hedged
for an immediate default, we are underhedged as the bond accretes to par.
However we are fully hedged if the bond price falls. On the other hand, the
carry becomes more negative for a face value hedge as the bond price falls
since we become overhedged to an immediate default. This over-hedging
reduces as the bond accretes to par.

We emphasise that Figure 4 shows the theoretical relationship between the carry
and the gain on default. In practice, other factors (Figure 6) can create disloca-
tions between the cash and default swap markets which can enable the investor to

Lehman Brothers, October 18, 2001 17

Quantitative Credit

Figure 5. Closing out a long protection default swap position by entering

into the offsetting transaction
Long Protection paying
Default Swap Spread D(0,T)

Long Protection paying

Default Swap Spread D(0,T) Close out by
entering into
offsetting
transaction 100-Recovery
following
credit event

100-Recovery
100-Recovery following
following credit event
credit event

Short Protection receiving

Default Swap Spread D(t,T)

either increase their carry or increase the gain in the event of a default.

One way to view a long basis (long cash, long protection) position is as a cov-
ered put option. The option's intrinsic value is 100%-Full Price which declines
if the credit improves or as the bond price pulls to par, and increases if the credit
deteriorates. The negative carry on this trade is equivalent to the amortised op-
tion premium. Note that as the default swap premium terminates on default, the
earlier the default, the less premium we will have paid for this option.

Default Swaps Used as a Spread Product

The default swap is essentially a spread product. This is shown most clearly in
the determination of the mark-to-market of the default swap, calculated as the
value of the initial trade plus the value of the offsetting transaction.

For example, an investor who has purchased protection can close out the trans-
action by selling protection on the same credit, to the original maturity date.
Exposure to the payment at default is then hedged out. However, the remaining
premium is only paid until maturity or default, whichever occurs sooner. If de-
fault happens, any remaining carry is lost. To mark the position to market, we
need to know the probability of surviving to each premium payment date. Using a
simple model1, we can write the mark-to-market at time t as
N
MTM(t) = ( D(t , T ) − D(0, T ) ) ∑ ∆i (b)Q(i ) Z (i )
i =1

1 We assume independence of the interest rate process, the default process and the recovery amount.
It is also assumed that recovery is paid as a fixed percentage of the facevalue of the bond. For
simplicity, we have also ignored the effect of the accrued premium paid at the time of default.

18 Lehman Brothers, October 18, 2001

 Figure 6. Reasons for the Basis

In reality, a number of technical and market-driven factors mean that the default swap
spread and the asset swap spread do not always trade at the same level, even when the
bond is priced at par. We set out some of these factors:

Why default swap spreads should exceed the asset swap spread
• Delivery Option - the protection buyer is long an option to choose one out of
a basket of deliverable assets to be delivered in the event of default
• Risk of Technical Default - default swaps may be triggered by events which
do not constitute a full default on the corresponding cash asset. A protection
seller may demand a higher spread to compensate them for this risk.
• Profit and Loss (P&L) realisation - unwinding a default swap by entering into
the offsetting transaction means that any P&L is only realised at maturity or
default whereas the P&L on a bond can be realised immediately by selling it.
• Liquidity in default swaps is focused at the 3 and 5 year maturities. Away
from these, bid-ask spreads are wider. They also widen on a credit deteriora-
tion.
• Demand for protection - the difficulty in shorting a credit in the cash market
makes default swaps the best alternative and so widens default swap
spreads more than cash spreads.

Why the default swap spread should be less than the asset swap spread
• Funding costs - we believe that most market participants fund above Libor.
For these investors, selling protection (which implies Libor funding) is cheaper
than buying the asset and so default swap spreads narrow.
• Counterparty credit risk - the protection buyer is also exposed to the credit
quality of the counterparty in the event of default.
• Market Short - the huge market for synthetic CDOs has resulted in an excess
of protection sellers in the default swap market.
• Liquidity in default swaps is often better than that for cash at the standard 3
and 5-year maturities when comparing the same notional sizes.
• In an asset swap, the asset buyer is exposed to an unknown mark-to-
market on the interest rate swap in the event of default. A wider asset swap
spread may be demanded as compensation.

where D(t,T) is the value at time t of the default swap spread to maturity date T,
D(0,T) is the initial value of the default swap to maturity date T at trade inception,
Q(i) is the probability of surviving to the ith premium payment date, Z(i) is the Libor
discount factor to the ith premium payment date and Di(b) is the year fraction for the
spread payment in the appropriate accrual basis b. The values of Q(i) can be extracted
from a model of default and recovery which is calibrated to market default swap
spreads. It is clear from the above expression for the mark-to-market that:

1) The mark-to-market of a long protection default swap position increases if the

default swap spread widens. This dependence is sub-linear since Q(i) is itself
a decreasing function of D(t,T). Similarly, we can show that the mark-to-mar-
ket of a short protection position falls as the default swap spread increases.

2) As the maturity of the trade shortens the number of remaining default swap
spread payments decreases and the mark-to-market declines. This is known
as the "theta" of the position.

Lehman Brothers, October 18, 2001 19

Quantitative Credit

Taking a View on changes in the Basis

Until now, we have simply discussed the theoretical relationship between cash
and default swaps. However, a number of additional factors, both technical and
market-driven, can cause cash and default swap markets to dislocate. These are
summarized in Figure 6. We now discuss some of the possible strategies that
may be implemented to take advantage of changes in the basis caused by dynamic
changes in these factors.

Investors can use the default swap basis to express a view on the delivery option
(see Figure 6). This option becomes more valuable for credits as the implied
probability of default increases. Consequently, the default swap spread widens
faster than the bond spread. To assess the value of the option, investors should be
aware as to what assets are deliverable into the contract: the type of bonds in
terms of their maturity, coupon, liquidity and any non-standard features, as well
as loans and other contingent liabilities2. If the investor's view is that the credit is
on the path to default, they can play the delivery option by buying default protec-
tion and buying the cheapest-to-deliver asset, thereby maximising their gain in
the event of default.

Investors can profit from a long basis position if the credit deteriorates. In this
scenario, both bond and default swap spreads widen and we generally find that the
bid-ask in the default swap market becomes wider than that in the cash market. As
the credit becomes more distressed, the increased likelihood of default, demand
for protection, and the fact that the seller of protection is short the delivery
option reduces the demand to sell protection, thus driving out default swap spreads.
Liquidity in default swaps reduces, leaving bonds, and in certain circumstances,
loans as the only instruments with any liquidity. In this situation the basis be-
tween bonds and default widens to a maximum level and the position may be
unwound at a profit.

An investor can profit from a long basis position when spreads tighten. In a spread-
tightening scenario, default swap spreads lag bond spreads as the value of the
delivery option is linked to the less liquid liabilities which react more slowly to
the credit improvement. This is reinforced by the fact that investors expressing a
positive view on the credit tend to do so in the cash market. Hence the bond price
increases by more than the decline in the mark-to-market of the protection re-
sulting in an overall gain. This will be further enhanced by the fact that the bid-ask
on the bond is tighter than when the trade was entered into. At some point the
default swap market will catch up with the bond market. In this case the investor
can profit from a short basis position as the default swap spreads tighten by more
than the cash.

If the credit quality of an issuer is not expected to change, the relationship be-
tween cash and default swaps would also not be expected to change. In this case,

2 Note that the recent ISDA supplement (May 2001) has altered the specification of the deliverable
assets.

20 Lehman Brothers, October 18, 2001

 the investor should look for positive carry opportunities where the trade can be
unwound at a later date at a cost lower than the accumulated carry.

Conclusion
In this article, we have explained the theoretical relationship between cash and
default swaps as well as describing some of the many factors which can cause
this relationship to break and how to implement default swap basis trades. We
believe that an understanding of all of these issues is vital for those wishing to
avail themselves of this new trading opportunity.

Lehman Brothers, October 18, 2001 21

Quantitative Credit

Dominic O’Kane LEVERAGING SPREAD PREMIA WITH DEFAULT BASKETS

Quantitative Research
[email protected] An investor can use default baskets to leverage spread premia. We show that
44-20-7260-2628 first and second-to-default baskets compare favourably to single name
default swaps on a risk-return basis.
Lutz Schloegl
Quantitative Research Introduction
[email protected] Credit spreads reflect the additional compensation that investors demand for
44-20-7601-0011, ext. 5016 holding defaultable rather than default-free securities. For this reason, investors
in corporate and certain emerging market government securities demand yields
that are higher rather than those of US (and other) government securities. These
are generally believed to be default-free. However it is well known in the credit
markets that the credit spread earned can be even greater than the expected loss
computed from an historical analysis of default rates for similarly rated issuers.
This excess spread, known more formally as a risk-premium, is understood to be
compensating investors for risk-aversion, market-to-market risk, the lower li-
quidity compared to government bonds, and the lack of full diversification in the
investor’s credit portfolio. It also incorporates the effect of other technical and
structural market factors.

The best way to take advantage of the risk and liquidity premia in credit markets
is through a diversified pool of credits. A large pool of well-diversified assets
minimises the investor’s exposure to the specific risk while enabling them to
enjoy the risk premium.

Default baskets are the simplest way by which investors can gain a leveraged
exposure to a pool of typically between 5 and 10 issuers. This article aims to
determine how the risk premium of basket default swaps differs from that of
single-name assets, and whether this provides opportunities for investors on a
risk-return basis.

Default baskets have been described elsewhere in significant detail (see

references). Briefly, a first-to-default basket is an agreement in which a
protection buyer purchases protection from a protection seller against the loss
made on the first asset which defaults in a specified basket of credit names. As
soon as default occurs the protection buyer delivers the defaulted asset to the
protection seller in return for par. In a second-to-default basket the payment is
triggered by the second default in a basket of specified issuers. Once again the
defaulted asset is delivered by the protection buyer who receives par. In both
cases, the protection buyer pays for the protection in the form of a fixed regular
spread which is paid until default or maturity, whichever occurs first. Clearly,
third to default and higher are obvious extensions. However, the size of the baskets,
typically between 5 and 10 issuers, means that 3 or more defaults have a very low
probability of occurring so these are generally low yielding assets which are of
limited interest to most investors. We therefore focus our analysis on first and
second-to-default baskets.

22 Lehman Brothers, October 18, 2001

 An important point to keep sight of in the following is the fact that a basket’s
leverage is determined by its order of protection in conjunction with the size of
the underlying pool. A second-to-default basket on a pool of twenty issuers will
have a larger likelihood of being triggered, and hence a much greater leverage
than one on a pool of five issuers. A second-to-default basket on the larger pool
behaves in a very similar fashion to a first-to-default basket.

Basket Spread Compensation

We want to compare the risk compensation an investor can receive on default
baskets with that which is available via individual issuer default swaps. In practice
it is difficult to disentangle the different factors contributing to default swap
spreads. In addition to the pure default risk component, credit spreads are influ-
enced by many other factors, such as liquidity, the risk of technical default, and
repo considerations.

One way to obtain an indication of the risk premium inherent in spreads is to

compare the market-implied default probabilities with average historical defaults.
In fact, we can also follow this path in the opposite direction; i.e. construct credit
curves from historical average default rates and compare the default swap spreads
which would be implied by such a curve with those observed in the market.

The spread we obtain from such a “historical” curve gives us a measure of the fair
compensation for the expected loss. In the language of financial theory, this is
the real-world spread and compensates the investor for the historical average
loss. The excess spread, defined as the (annualised) spread minus the real-world
spread, even though it is a product of many factors, gives us a measure of the
extra premium investors receive for holding this credit risk in default swap form.

Since higher spreads are usually associated with higher risk, what we also need is
a metric that tells us about how much compensation we are receiving per unit of
risk. We therefore introduce the ratio of the default spread s paid by the market
to the spread s% implied from historical default rates, which we call the spread
coverage ratio. To see why this is an informative number, note that the expected
premium that will be collected over the life of a trade, and which therefore de-
pends upon the time of default, in a default swap that pays a spread of s is given by

E [ Prem ] = s A (1)

where A is the value of a risky annuity i.e. it terminates at default or maturity, and
the present-value is computed with historical default rates. The break-even con-
dition for the historical spread is given by

E [ Loss ] = s% A (2)

Therefore
E [ Prem ] s
=
E [ Loss ] s% (3)

Lehman Brothers, October 18, 2001 23

Quantitative Credit

The spread coverage ratio tells us how many times the expected basket premium
covers the expected loss. Though both measures (excess spread and spread cov-
erage ratio) clearly only approximate the amount of risk compensation an investor
receives, they do allow us to make a relative value comparison of default bas-
kets and plain vanilla default swaps.

Starting with a simple model for Basket Spreads

To gain an insight into how default baskets leverage the excess spread of the
individual assets we start with a simple model. We take a basket with just two
issuers A and B. The maturity of the basket is assumed to be T years, and the risk-
neutral probability of A defaulting before maturity is given by PA while the
risk-neutral probability of B defaulting is given by PB . The probability of both
assets defaulting within the life of the trade is given by PAB . Assuming that both
assets have the same recovery rate, R, we can write the risk-neutral value of the
first-to-default and second-to-default spreads as

S FTD = (1 − R) ( PA + PB − PAB ) (4)

and

S STD = (1 − R) PAB (5)

Without compromising this analysis we can, for simplicity, ignore the time value
of the payment of par minus the recovery rate upon the triggering of the default
- i.e. we effectively assume interest rates of zero.

We can also calculate the expected loss in the real-world measure where we
denote the analogous real-world default probabilities to those defined above as
P%A , P%B and P%AB . Likewise we define the real-world values of the protection for
first and second-to-default baskets as

(
S%FTD = (1 − R ) P%A + P%B − P%AB ) (6)

and

S%STD = (1 − R) P%AB (7)

Hence, the first-to-default coverage ratio is given by the risk-neutral spread di-
vided by the value of the real-world spread, i.e.

S FTD PA + PB − PAB
CFTD = = (8)
S%FTD P%A + P%B − P%AB

since the recovery rate dependency cancels. Likewise the second-to-default cov-
erage ratio is given by

24 Lehman Brothers, October 18, 2001

 S STD PAB
CSTD = = (9)
S%STD P%AB

To see how this depends upon default correlation, we need to specify a way of
modelling correlated defaults. One choice gaining favor in the credit derivatives
market is to use a Merton-style model of correlated default in which default
occurs when the asset value of a firm falls below a certain threshold. In such a
model the default correlation arises through the correlation of the asset values
of the various issuers. The correlation structure is defined using a specific choice
of copula function. A copula is a function which specifies how the joint default
dependence structure depends on the marginal default distributions of the indi-
vidual issuers. We can therefore write the joint default probability as a function
of the individual default probabilities and the parameters of the joint distribution
of the asset values, i.e.

PAB = Φ( PA , PB , Γ) (10)

where Φ ( x, y , Γ ) is the cumulative density function and G is a vector of param-

eters, including the asset value correlation, which depends on the type of copula
function used. In particular, choosing a normal copula implies that Φ ( x, y , ρ )
is the cdf of the bivariate normal distribution with correlation coefficient ρ .

Within the Merton-style framework, asset values are represented by diffusion

processes, and their instantaneous correlation is the same in the real-world mea-
sure as it is in the risk-neutral measure. We can therefore write

P%AB = Φ ( P%A , P%B , ρ ) (11)

For reasons described above, it is generally found that the risk-neutral single-
name default probabilities exceed the real-world default probabilities by an amount
that consists of risk premia and other components. Let us therefore assume that
the ratio of the risk-neutral to real-world default probabilities is given by k, where
k>1, and that this is the same for both assets, i.e. PA = kP%A and PB = kP%B . We
can then write

CFTD =
( )
k P%A + P%B − Φ (kP%A , kP%B , ρ )
(12)
P%A + P%B − Φ( P%A , P%B , ρ )

At zero correlation we have Φ ( P%A , P%B ,0) = P%A P%B and Φ ( kP%A , kP%B ,0) = k P%A P%B .
2

These terms are small and so, to first order, the coverage ratio for a first-to-default is
approximately equal to k. In the limit of high correlation we have
Φ( P%A , P%B , ρ ) = Min[ P%A , P%B ] , and Φ(kP%A , kP%B , ρ ) = k Min[ P%A , P%B ] so that
CFTD = k exactly. Hence the coverage ratio for a first-to-default is very close to k for
all values of correlation.

Lehman Brothers, October 18, 2001 25

Quantitative Credit

For second-to-default baskets the coverage ratio is given by

P%AB Φ (kP%A , kP%B , ρ )

CSTD = = (13)
PAB Φ ( P%A , P%B , ρ )

This has a significant dependence on the correlation between asset A and asset B.
Let us take the case of zero correlation. In this case,

Φ(kP%A , kP%B , 0) = k 2 P%A P%B (14)

and
Φ( P%A , P%B , 0) = P%A P%B (15)

so that the spread coverage ratio is then given by

CSTD ( ρ = 0) = k 2 (16)

What is happening is that the spread coverage ratio of each asset in a second-to-
default is being squared because at zero correlation the probability of two defaults
is the product of the two default probabilities. This effect significantly enhances
the coverage ratio of a second-to-default. In the other limit of maximum corre-
lation, both assets will tend to default together with a probability equal
to Min[ PA , PB ] so that the second-to-default is effectively a first-to-default and
we get

CSTD ( ρ = ρ MAX ) = k (17)

Between these two limits we can plot the coverage ratio as a function of the asset
correlation as shown below in Figure 1.

Figure 1. Spread coverage ratios for first and second to default baskets on A
and B where k=5 and Pa=Pb=10%.

30
Spread Coverage Ratio

25

20

15

10

0
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
Correlation
Spread Coverage Ratio for First to Default
Spread Coverage Ratio for Second to Default

26 Lehman Brothers, October 18, 2001

 We see clearly that the spread coverage ratio for the first to default basket is a
very flat line at a value very close to 5, the chosen value for k. This agrees with
the analysis above. We also see that the spread coverage ratio for the second-to-
default basket begins at a value of k2=25 for low correlation, and then falls to
k=5 as the correlation increases.

For more assets we need to know about the probability of all of possible joint
default events. We need to move to a proper model of correlated default that
should also take into account the effect of interest rates and the term structure
of credit spreads. We do this in the next section. However, despite the simplistic
nature of this analysis, the results derived here apply just as much as they do to
the more sophisticated model used in the next section.

A General Model of Correlated Default

Clearly the most important driver of basket spreads is the credit quality of the
underlying assets. Besides this, the next most important element is the degree of
dependence that is assumed between the individual issuers. What we need is a
statistical model which allows us to assign probabilities to joint default events
given the individual default probabilities.

An elegant and increasingly popular way of modelling the joint dependency of

the default times of the issuers in a basket is via copula functions. The advantage
of copula functions is that they allow us to disentangle the modelling of the
dependence structure of the different default times from that of their indi-
vidual distributions. For any choice of copula function, we obtain a valid model
for the joint distribution of the issuer default times that allows us to evaluate a
default basket. Popular copula functions include the normal and the student-t
copula, they are treated in detail in the article by Mashal and Naldi, also in this
issue. Once we have chosen a copula function, we can easily evaluate default
baskets by generating issuer default times in a Monte Carlo simulation.

Since we know the times of defaults, we can easily take into account the present
valuing of any cashflows. We can then work out the breakeven basket spread as
the spread which equates the present values of the premium and protection legs,
taking into account the fact that the spread leg terminates once the basket has
been triggered.

Rating Based Analysis

Using a model of the type described in the previous section, we can analyse the
pricing of portfolios of two or more assets, taking into account a full dependency
structure as well as the effect of interest rates. In what follows, we consider collat-
eral with a rating of A3. On average, 5-year protection on an issuer with this rating
will be trading around 130 bp in the default market. For our comparison, we use the
average default rates per rating as supplied by Moody’s, c.f. Table 1.

Based on a credit curve constructed from these historical default rates, and as-
suming an average recovery rate of 45%, the 5-year default swap spread for an

Lehman Brothers, October 18, 2001 27

Quantitative Credit

A3 rated issuer would be 4 bp. In other words, for A3 rated credits, the default
swap market is on average demanding an excess spread of 130-4=126 bp and a
spread coverage ratio of 130/4=32.5. It is important to note that both of these
numbers must be seen as a function of the issuer’s credit quality. The spread
coverage ratio in particular declines markedly as we go down the rating scale.

We consider a default basket consisting of five A3 rated issuers. The average

default swap spread equals 130bp. Using a model for the issuer dependence such
as we have described above, we can calculate the first and second-to-default
spreads as a function of the issuer correlation. This is assumed to be the same
for all issuer pairs. The basket spreads using market inputs are shown in Figure 2,
while Figure 3 shows the basket spreads obtained if we use issuer spreads based
on historical default rates.

In both cases, we note the typical correlation dependence of the basket spreads.
The first-to-default (FTD) spread is decreasing in the issuer-issuer correlation -
as correlation increases, joint events become more likely as does the probability
of all of the assets surviving. On the other hand, the second-to-default (STD)
spread is only mildly dependent on it, starting at a low spread at low correlations
and increasing with correlation as joint defaults become more likely. In our com-
parison of baskets with individual assets, we use a correlation of 60%, which we
regard as being conservative.

Using an issuer correlation of 60%, the FTD spread based on market spreads is
375 bp, whereas the spread based on historical default rates is 15 bp. This gives
an excess spread of 375-15=360 bp and a ratio of 375/12=25. To compare this
to a single-name default swap, we must look to an asset that pays a spread of 375
bp. We estimate that an issuer paying this amount of spread should be rated around

Table 1. Average cumulative default rates by rating from 1 to 10 years for the period 1983 – 2000

Rating 1 2 3 4 5 6 7 8 9 10
Aaa 0.00% 0.00% 0.00% 0.06% 0.18% 0.25% 0.34% 0.43% 0.43% 0.43%
Aa1 0.00% 0.00% 0.00% 0.21% 0.21% 0.35% 0.35% 0.35% 0.35% 0.35%
Aa2 0.00% 0.00% 0.06% 0.18% 0.41% 0.49% 0.59% 0.71% 0.85% 1.01%
Aa3 0.06% 0.09% 0.17% 0.26% 0.37% 0.49% 0.49% 0.49% 0.49% 0.49%
A1 0.00% 0.03% 0.30% 0.47% 0.59% 0.73% 0.79% 0.86% 0.86% 0.96%
A2 0.00% 0.02% 0.16% 0.41% 0.62% 0.84% 0.99% 1.35% 1.63% 1.71%
A3 0.00% 0.12% 0.22% 0.30% 0.35% 0.47% 0.68% 0.77% 0.97% 1.09%
Baa1 0.07% 0.30% 0.53% 0.86% 1.19% 1.43% 1.82% 2.05% 2.20% 2.20%
Baa2 0.06% 0.29% 0.61% 1.22% 1.89% 2.54% 2.93% 3.17% 3.46% 3.81%
Baa3 0.39% 1.05% 1.62% 2.47% 3.15% 4.09% 4.99% 5.95% 6.54% 7.03%
Ba1 0.64% 2.10% 3.81% 6.15% 8.12% 10.09% 11.43% 12.75% 13.35% 14.08%
Ba2 0.54% 2.44% 4.95% 7.32% 9.27% 10.88% 12.59% 13.60% 14.27% 14.71%
Ba3 2.47% 6.82% 11.68% 16.18% 20.63% 24.74% 28.39% 32.28% 35.83% 38.22%
B1 3.48% 9.71% 15.59% 20.56% 25.62% 30.78% 36.15% 40.30% 44.16% 48.01%
B2 6.23% 13.70% 20.03% 24.63% 28.24% 31.14% 32.73% 34.33% 35.03% 35.90%
B3 11.88% 20.18% 26.71% 31.95% 36.68% 39.89% 42.81% 46.80% 51.42% 53.53%
Caa1-C 18.85% 28.29% 34.51% 40.23% 43.42% 46.48% 46.48% 49.73% 53.92% 59.04%

Source: Moody’s Investor Services

28 Lehman Brothers, October 18, 2001

 Figure 2. Basket spreads for five A3 issuers with market spread inputs

500
450
400
350

Spread (bp)
300
250
200
150
100
50
0
45% 49% 52% 56% 59% 63% 67% 70% 74% 77% 81% 85% 88%
Correlation
FTD STD

Baa3. Using historical default statistics, such an issuer provides an expected loss
of 35bp. The excess spread for the single-name is therefore 375-35=340 bp, and
the spread coverage ratio is only 375/35=11. Therefore, in order to earn the
same spread as from the first-to-default basket on investment grade names,
the investor has to take exposure to a sub-investment grade asset, and fur-
thermore, this asset offers a significantly lower spread coverage ratio as
well as less excess spread.

According to the data in Table 1, an issuer with the same real-world spread of 15
bp as the basket would be rated Baa1. On average, default protection on this type
of issuer trades around 200 bp, giving an excess spread of 200-15=185 bp and a
spread coverage ratio of 200/15=13.3. We see that the basket dominates the
single asset in both metrics.

Let us now turn to the second-to-default basket. Again using an issuer corre-
lation of 60%, the market STD spread is 168 bp, while the STD spread based
on historical default rates is merely 3bp. This gives an excess spread of 168-
3=165 bp and a spread coverage ratio of 168/3=56. The STD basket clearly
benefits from the diversification in the issuer pool. It pays a slightly higher
spread than the individual issuers in the pool, while offering a greater cover-
age ratio. The size difference between the coverage ratio of the
second-to-default basket and an individual asset must be taken with caution,
because the real-world spreads are so low. However, the second-to-default
basket position has the additional safety cushion that at least two assets have
to default before there is a loss.

In other words, FTD baskets are suited to investors trying to maximize the
earned spread for a given level of coverage ratio, while STD baskets are suited
to investors looking to earn an investment grade spread while maximizing
their coverage ratio.

Lehman Brothers, October 18, 2001 29

Quantitative Credit

Figure 3. Basket spreads for five A3 issuers with spread inputs based on
historical default rates

18
16
14
12

Spread (bp)
10
8
6
4
2
0
45% 49% 52% 56% 59% 63% 67% 70% 74% 77% 81% 85% 88%
Correlation
FTD STD

As the basket spread is a function of the spreads of the issuers in the pool, the
obvious way to increase the spreads that can be earned is to have assets of lower
credit quality in the pool. Alternatively, we can avoid exposure to sub-investment
grade assets by increasing the leverage of the basket positions. This is achieved
by increasing the size of the issuer pool that the basket payoffs reference. We
therefore consider a pool of ten A3 rated issuers. The basket spreads are shown
in Figure 6 and Figure 7, while the spread ratios are shown in Figure 8. Clearly,
the larger basket means that the spreads are significantly higher than in the five-
issuer case. As before, we examine the 60% issuer correlation case. The FTD
spread using market inputs is 531 bp, while the spread based on historical default
rates is 26 bp. This gives an excess spread of 531-26=505 bp and a spread cover-
age ratio of 531/26=20. The FTD position on a basket of ten A3 issuers pays a

Figure 4. Ratio of basket spreads for five A3 issuers

100
Spread Coverage Ratio

80

60

40

20

0
45% 49% 52% 56% 59% 63% 67% 70% 74% 77% 81% 85% 88%
Correlation
FTD STD

30 Lehman Brothers, October 18, 2001

 Figure 5. A comparison of a first-to-default basket with an asset which pays
the same spread and one which has the same expected loss

FIRST TO DEFAULT

Basket of 5 A3 issuers paying an

average spread of 130bp with an
equity correlation of 60%

FTD Basket Spread = 375bp

Excess Spread = 360bp
Coverage Ratio = 25
SAME EXPECTED
SAME SPREAD LOSS

Single name Baa3 issuer

Single name Baa1 issuer
Issuer spread = 375bp
Issuer spread = 200bp
Excess spread = 340bp
Excess spread = 175bp
Coverage ratio = 11
Coverage ratio = 13.3

high-yield type spread while still offering a better spread coverage ratio than the
Baa3 asset we examined earlier.

The STD spreads are 295 bp using market inputs and 7 bp based on historical
default rates, giving an excess spread of 295-7=288 bp and a spread coverage
ratio of 295/7=42. We see that once again, for the spread it pays, the STD basket
position maximizes the coverage ratio.

Figure 6. Basket spreads for ten A3 issuers with market spread inputs

700
600
500
Spread (bp)

400
300
200
100
0
45% 49% 52% 56% 59% 63% 67% 70% 74% 77% 81% 85% 88%
Correlation
FTD STD

Lehman Brothers, October 18, 2001 31

Quantitative Credit

Figure 7. Basket spreads for ten A3 issuers with spread inputs based on
historical default rates

35
30
25

Spread (bp)
20
15
10
5
0
45% 49% 52% 56% 59% 63% 67% 70% 74% 77% 81% 85% 88%
Correlation
FTD STD

Conclusions
Baskets provide a third way to leverage credit risk premia - the other two ways
being to drop down the credit spectrum and so increase the likelihood of de-
fault, or to drop down the capital structure and so to risk losing more in the
event of a default. First-to-default baskets leverage credit risk premia by ex-
posing the investor to the first default in a pool of credits. While this has the
effect of increasing the likelihood of a default event, it does so by exposing
the investor to liquid issuers which are typically investment grade quality and
which are well-known to the credit investor. Indeed the customisable nature of
baskets means that an investor has the ability to select exactly which credit
names are used. It also allows the investor to express a view about the default
correlation between the assets in the pool.

Figure 8. Spread coverage ratio of basket spreads for ten A3 issuers

70
Spread Coverage Ratio

60
50
40
30
20
10
0
45% 49% 52% 56% 59% 63% 67% 70% 74% 77% 81% 85% 88%
Correlation
FTD STD

32 Lehman Brothers, October 18, 2001

 In this piece we have shown that default baskets, be they first or second-to-de-
fault, provide new ways for investors to leverage the credit risk premia embedded
within credit spreads. Whether first or second-to-default should be chosen de-
pends very much on the investor’s risk appetite. Those seeking high spreads should
find first-to-default baskets superior to single name assets since they leverage
the excess spread more than single-name assets with the same spread. For more
risk-averse investors, second-to-default baskets maximise the ratio of spread
paid per unit of expected loss when compared to single name assets.

Reference
[1] Modelling Credit: Theory and Practice, Dominic O’Kane and Lutz Schloegl,
Lehman Publications, February 2001
[2] Credit Derivatives Explained, Dominic O’Kane, Lehman Publications,
March 2001
[3] Basket Default Swaps and the Credit Cycle, Marco Naldi, Quantitative Credit
Research Quarterly, June 30, 2001
[4] Extreme Events and the Valuation of Multi-name Credit Derivatives, Marco Naldi
and Roy Mashal, This issue of the Quantitative Credit Research Quarterly

Lehman Brothers, October 18, 2001 33

Quantitative Credit

EXTREME EVENTS AND THE VALUATION OF DEFAULT BASKETS

Marco Naldi Introduction

[email protected]
Specifying an appropriate model for dependent defaults is the core problem in
Roy Mashal the valuation of multi-name credit derivatives. Different dependence structures
produce different default distributions, which in turn affect the valuation of prod-
ucts such as CDO tranches and basket default swaps.

Several credit models rely on Merton’s (1974) idea that a firm defaults when its
asset value drops below its liabilities. A straightforward extension of this frame-
work to the multivariate case implies that the dependence structure of defaults is
determined by the assumed joint distribution of asset returns. Many existing
portfolio models assume multivariate normality as the underlying joint distribu-
tion of asset returns, even if the normal distribution is clearly not compatible
with the extreme joint realizations that we observe in the data. The market tur-
moil spurred by the tragic events of September 11 is reminding us once more
that extreme joint movements do happen more often than the normal distribution
would predict.

In this article we introduce POPSTAR (POrtfolio Pricing Simulator with TAil

Relations), a simulation-based pricing model which accounts for extreme events
by means of a more realistic distributional assumption. After describing the prop-
erties of its simulation engine, we discuss how to make the model operational,
with particular reference to the estimation of the parameters that determine the
likelihood of extreme events. Finally, we show the pricing impact of taking ex-
treme events into consideration. For the purpose of this discussion, we focus
our attention on basket default swaps, since the valuation of these contracts is
particularly sensitive to the dependence structure of defaults.

Figure 1. Nth-to-Default Basket Swap

Pays Premium

Protection Buyer Protection Seller

Pays at the Nth default

A,B,C,…
Reference Names

1 For more on Nth-to-default swaps and other credit derivatives, see O’Kane and Schloegl (2001).

34 Lehman Brothers, October 18, 2001

 In an Nth-to-default basket swap, two counterparties agree on a maturity and a set
of reference assets, and enter into a contract whereby the protection seller peri-
odically receives a premium (also called "basket spread") from the protection
buyer. In exchange, the protection seller stands ready to pay the protection buyer
par minus recovery of the Nth referenced defaulter in the event that the Nth de-
fault occurs before the agreed-upon maturity. First- and second-to-default swaps
are the most popular orders of protection.1

Default Correlation and the Distribution of Asset Returns

Default correlation measures the tendency of two credits to default jointly within
a specified horizon. Formally, it is defined as the correlation between two binary
random variables that indicate defaults, i.e.

p AB − p A p B
ρD = (1)
p A (1 − p A ) p B (1 − p B )

where pA and pB are the marginal default probabilities for credits A and B, and
pAB is the joint default probability. Of course, pA , pB and pAB all refer to a
specific horizon. Notice that default correlation increases linearly with the joint
probability of default and is equal to zero if and only if the two default events are
independent.

Default correlations are the fundamental drivers in the valuation of multi-

name credit derivatives. Unfortunately, the scarcity of default data makes joint
default probabilities, and thus default correlations, very hard to estimate directly.
As a result, researchers have developed alternative methods to calibrate the fre-
quency of joint defaults within their valuation models.

One way to simulate correlated defaults relies on the use of copula functions.
Generally speaking, copulas are used to link marginal and joint distribution func-
tions. Starting with an assumption for the marginal distributions of default times,
a copula function can be employed to obtain their joint distribution. This can
then be used as the probability law underlying a very efficient time-to-default
simulation. This procedure is extremely useful for the valuation of multi-issuer
credit derivatives, since we can extract (risk-neutral) marginal default probabili-
ties from liquid single-name products, and then value multi-name contracts by
simulating correlated default times.

To illustrate this point, consider two credits A and B, whose default times TA and
TB are exponentially distributed with hazard rates hA and hB. A joint distribution
that correlates TA and TB while respecting their marginals can be obtained by
means of a bivariate normal copula

FN ( x, y ) = P(T A < x, TB < y) = −1 −1 (2)

2 ( (E hA (x)), (E hB (y)),r),

Lehman Brothers, October 18, 2001 35

Quantitative Credit

where F2(× ,× , r) denotes a bivariate standard normal distribution with correlation

r, F(× ) is a univariate standard normal distribution, and Eh( × ) is an exponential
distribution with hazard rate h. Taking limits, it is straightforward to verify that
this joint distribution is perfectly legitimate in that it respects the exponential
marginals that we started with.

Li (2000) and Nyfeler (2000) show that several existing portfolio models gen-
erate a dependent structure of defaults that is in fact equivalent to the one
produced by a normal copula. Notice that the framework described by equation
(2) can be reinterpreted as a variation of Merton’s model where:

1) default is caused by a firm's normally distributed asset return falling below

the threshold F -1(Eh( × )), and
2) the parameter r represents the correlation between the jointly normal asset
returns of the obligors.

This structural interpretation is useful for two reasons. First, it simplifies cali-
bration by relating the copula parameter r to the correlation between asset returns.
This offers a clear advantage since reasonable proxies for asset correlations can
be computed from observable equity data. Second, it reveals that the choice of a
particular copula for survival times may be related to an implicit distributional
assumption for asset returns.

The assumption of normality of asset returns is certainly not innocuous, since a mul-
tivariate normal distribution does not allow for extreme joint events to happen with
the frequency that the data suggest. Figure 2 shows a bivariate scatterplot of stan-
dardized equity returns using 7 years of monthly data (August 94 - July 01). To the
extent that equity returns proxy for asset returns, this figure highlights the major
problem with the normality assumption. According to the normal distribution, the

Figure 2. Standardized Monthly Returns

1
Wal-Mart

-1

-2

-3

-4
-4 -3 -2 -1 0 1 2 3 4
Honeywell

36 Lehman Brothers, October 18, 2001

 most extreme joint realizations in this plot have a likelihood of happening of the
order of one in a hundred thousand; yet, we observe them in a sample of 84 points.
The ability of a multivariate distribution to accommodate joint extreme events can
be related to the concept of “tail dependence”. Formally, for two random variables
X and Y with marginal distributions FX and FY, (lower) tail dependence is defined
as:

(
λ = lim P Y < FY−1 (u ) X < FX−1 (u ) .
u →0+
) (3)

In words, tail dependence measures the probability that Y will have a realization
in the tail of its distribution given that X has had a realization in the tail of its
own. The problem with the multivariate normal distribution is that l is identi-
cally equal to zero. Another manifestation of the same problem can be seen in
the thinness of the tails of a multivariate normal density, which implies that there
is very little probability mass on extreme joint events.

POPSTAR allows for “fat-tailed” returns by means of a non-normal assumption.

It does so by joining the marginal distributions of default times through a copula
function that introduces extreme events while retaining the structural interpreta-
tion described above. The latter point is important for practical purposes because
it permits relating it to the copula parameters to the distribution of asset returns.

More precisely, POPSTAR obtains the joint distribution of default times by

joining the exponential marginals Eh(× ) with the t-copula

(4)
Ft ( x, y ) = P(T A < x, TB < y) = t 2 ,v (t v-1(E h(x)),t v-1(E h(y)),r),

where t2,v(× ,× , r) is a bivariate standard t distribution with v degrees of freedom

and correlation r , and tv(× ) is a univariate standard t distribution with v degrees
of freedom.

One can immediately see that the structural interpretation of default is main-
tained. In fact, the choice of a t-copula for default times may be viewed as an
implicit assumption that asset returns follow a multivariate t distribution.
Since two jointly t2,v variables are marginally tv distributed, one can immedi-
ately interpret tv -1(Eh(× )) as a default threshold and t2,v(× ,× , r) as the joint
distribution of asset returns.

Not all copula functions lend themselves to a structural interpretation. Lindskog

(2000) and Schonbucher and Schubert (2001), among others, study different forms
of dependence such as Clayton and Gumbel copulas. These are perfectly legiti-
mate ways to join the marginal distributions of default times and introduce tail
dependencies. However, from a practical point of view, it is not clear where to
look for reasonable estimates of the copula parameters.

Lehman Brothers, October 18, 2001 37

Quantitative Credit

There is an impressive amount of literature documenting the non-normality of

equity returns (see for example Blattberg and Gonedes (1974)). When modeling
the joint distribution of equity returns, a multivariate t fits the data much better
than a multivariate normal. We will formally show this in the next section. Com-
pared with a normal, a t distribution has an additional parameter, the number of
degrees of freedom. As this number goes to infinity, a t distribution tends to a
normal distribution; i.e., it displays no tail-dependence. But for a finite number
of degrees of freedom, a t distribution allows for extreme joint realizations.

Figure 3 shows the values of tail dependence l for a bivariate t distribution as a

function of the correlation coefficient r and the degrees of freedom n. The last
row refers to the normal case (n=¥). To understand the meaning of these num-
bers, consider the case where the distribution is characterized by 10 degrees of
freedom and 30% correlation. In this scenario, knowing that the return of the
first name has had an extremely negative realization leaves us with a 3.31% prob-
ability that the return of the second name will also have an extremely negative
realization. This is very different from the prediction of a jointly normal distri-
bution. In the last section of this article, we will quantify the pricing impact of
this difference.

Figure 3. Tail Dependence (l)

ν/ρ -0.5 0 0.3 0.5 0.9 1

3 0.0257 0.1161 0.2161 0.3125 0.6701 1
10 0.0001 0.0068 0.0331 0.0819 0.4627 1
15 3e-6 0.0010 0.0097 0.0346 0.3724 1
20 9e-8 0.0002 0.0029 0.0151 0.3051 1
∞ 0 0 0 0 0 1

The relation between the tail dependence of asset returns and the dependence of
default events is depicted in figure 4. Using a 5-year horizon and two credits with
constant hazard rates of 1%, this graph compares a normal copula and a t-copula
with 3 and 10 degrees of freedom. Tail dependence increases default correlation
for any value of asset correlation. In particular, notice that even when asset re-
turns are uncorrelated (i.e. linearly independent), tail dependence can produce a
significant amount of default correlation.

Estimation and Testing

To simulate default times from the joint distribution specified in equation (4),
we need to estimate its parameters. Using time series of equity returns to proxy
for asset returns, we can estimate these parameters by maximum likelihood.
Moreover, since a multivariate t distribution tends to a normal distribution as the
number of degrees of freedom increases (i.e. Ft ® FN as n ® ¥), the normality

38 Lehman Brothers, October 18, 2001

 Figure 4. Correlation

0.8

Default Correlation
0.6 v=3
v=10
0.4 v=infinity

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Asset Correlation

Figure 5. Basket Examples

Basket Reference Names

#1 Fleet Boston, AT&T, IBM, Eastman Kodak.
#2 Honeywell, BellSouth, Wal-Mart, First Data
#3 Union of the names in baskets 1 and 2

assumption is actually nested in our framework as a special case. Therefore, we

can carry out likelihood ratio tests for the null hypothesis that returns are jointly
normal. For the analysis in this section and the next, we consider the baskets
described in figure 5.

Using 7 years of monthly data (Aug 94 – Jul 01) and assuming serial indepen-
dence of returns, figures 6-8 plot the likelihood of the multivariate return
processes for these three baskets as a function of the degrees of freedom. The
maximum likelihood estimates are shown to be 9, 7 and 9, respectively, which
signal the presence of a significant amount of tail fatness in the data. To confirm
the inadequacy of the normal distribution, the same figures also report the p-
values of the likelihood ratio tests for the null hypothesis of normality (n = ¥).
They suggest that we can reject normality with an infinitesimal probability of
making a mistake. This result is by no means specific to the chosen examples:
we have obtained analogous results with dozens of different portfolios.

Extreme Events and Valuation

Taking extreme events into account has significant consequences on the valua-
tion of multi-name credit derivatives. Other things equal, simulating defaults by
means of a fat-tailed copula increases the probability of joint defaults and, there-
fore, default correlations as defined in equation (1).

Lehman Brothers, October 18, 2001 39

Quantitative Credit

Figure 6. Basket #1 Log-Likelihood Function

365

360
ML Estimate

Function Value
DoF=9
355

350
LR test for Null=Normality,
p-value=1.33e-4
345

340
3 5 7 9 11 13 15 17 19 21 23 25 27 29 Normal
Degrees of Freedom

Figure 7. Basket #2 Log-Likelihood Function

370

367
ML Estimate
Function Value

DoF=7
364

361
LR test for Null=Normality,
p-value=2.34e-5
358

355
3 5 7 9 11 13 15 17 19 21 23 25 27 29 Normal
Degrees of Freedom

Figure 8. Basket #3 Log-Likelihood Function

760

755

750 ML Estimate
Function Value

DoF=9
745

740

735 LR test for Null=Normality,

p-value=5.55e-9
730

725
3 5 7 9 11 13 15 17 19 21 23 25 27 29 Normal
Degrees of Freedom

40 Lehman Brothers, October 18, 2001

 Figure 9 reports pricing results for first- and second-to-default swaps on the
baskets described in figure 5. All deals have a maturity equal to 3 years and have
been priced using information (LIBOR curve, market-implied hazard rates, esti-
mated parameters of t copula) available as of the end of July 2001. Premia are in
basis point and percentage standard errors are reported in brackets.

Figure 9. Valuation of 3-year Default Baskets (as of the end of July 2001):
Normality vs. POPSTAR

Basket Premium (StdErr)

1st-to-default 2nd-to-default
Normal 236 (0.28%) 29 (0.75%)
#1 POPSTAR 224 (0.26%) 36 (0.67%)
Diff -5% 24%

Normal 140 (0.34%) 12 (1.15%)

#2 POPSTAR 129 (0.35%) 19 (0.92%)
Diff -8% 58%

Normal 350 (0.29%) 64 (0.71%)

#3 POPSTAR 316 (0.31%) 76 (0.66%)
Diff -10% 19%

Recall from the previous section that accounting for extreme events increases
default correlations. As we showed in a previous issue of this Quarterly (see
Naldi (2001)), the sign of the relation between basket premia and default corre-
lations depends on the order of the basket. The value of first-to-default protection
is always monotonically decreasing in default correlations. Therefore, when we
allow for joint extreme events, first-to-default protection gets cheaper. The value
of second-to-default protection is not necessarily monotonic in default correla-
tions. Rather, it generally increases up to a certain point, then it becomes
decreasing. The location of this turning point depends on all other parameters
and, in particular, on the number of names in the basket. With a low number of
names, second-to-default protection is generally increasing in default correla-
tions over most of the domain. Intuitively, with only a handful of names in the
portfolio, the only way that two of them can default within 3 years is if they have
a significant tendency to default together. That explains the results we observe in
figure 9 regarding second-to-default valuation.

Lehman Brothers, October 18, 2001 41

Quantitative Credit

References

Blattberg, Robert C. and Nicholas J. Gonedes, 1974, “A Comparison of the Stable

and Student Distributions as Statistical Models for Stock Prices”, Journal
of Business, 47.

Li, David X., 2000, “On Default Correlation: A Copula Function Approach.”, Jour-
nal of Fixed Income, 9.

Lindskog, Filip, 2000, “Modelling Dependence with Copulas and Applications to

Risk Management”, Master Thesis, Swiss Federal Institute of Technology.

Merton, Robert C., 1974, "On the Pricing of Corporate Debt: The Risk Structure
of Interest Rates", Journal of Finance, 29.

Naldi, Marco, 2001, “Basket Default Swaps and the Credit Cycle”, Quantitative
Credit Research Quarterly (June), Lehman Brothers.

Nyfeler, Mark A., 2000, “Modelling Dependencies in Credit Risk Management”,

Diploma Thesis, Swiss Federal Institute of Technology.

O’Kane, Dominic and Lutz Schloegl, 2001, “Multi-Issuer Models of Default”,

Quantitative Credit Research Quarterly (June), Lehman Brothers.

Schonbucher, Philipp J. and Dirk Schubert, 2001, “Copula-Dependent Default

Risk in Intensity Models”, Working Paper, Department of Statistics, Bonn
University.

42 Lehman Brothers, October 18, 2001

 Publications—L. Pindyck, A. DiTizio, B. Davenport, W. Lee, D. Kramer, S. Bryant, J. Threadgill, R. Madison, A. Acevedo, K. Kim
This document is for information purposes only. No part of this document may be reproduced in any manner without the written permission of Lehman
Brothers Inc. Under no circumstances should it be used or considered as an offer to sell or a solicitation of any offer to buy the securities or other instruments
mentioned in it. We do not represent that this information is accurate or complete and it should not be relied upon as such. Opinions expressed herein are
subject to change without notice. The products mentioned in this document may not be eligible for sale in some states or countries, nor suitable for all types
of investors; their value and the income they produce may fluctuate and/or be adversely affected by exchange rates, interest rates or other factors.

Lehman Brothers Inc. and/or its affiliated companies may make a market or deal as principal in the securities mentioned in this document or in options or
other derivative instruments based thereon. In addition, Lehman Brothers Inc., its affiliated companies, shareholders, directors, officers and/or employees,
may from time to time have long or short positions in such securities or in options, futures or other derivative instruments based thereon. One or more
directors, officers and/or employees of Lehman Brothers Inc. or its affiliated companies may be a director of the issuer of the securities mentioned in this
document. Lehman Brothers Inc. or its predecessors and/or its affiliated companies may have managed or co-managed a public offering of or acted as initial
purchaser or placement agent for a private placement of any of the securities of any issuer mentioned in this document within the last three years, or may, from
time to time perform investment banking or other services for, or solicit investment banking or other business from any company mentioned in this document.
This document has also been prepared on behalf of Lehman Brothers International (Europe), which is regulated by the SFA. ©2001 Lehman Brothers Inc. All
rights reserved. Member SIPC.