Understanding credit curves for

multi-reference credit derivatives

Theodor Munteanu
10 October 2020

Contents
1 Basket default swaps 1

1 Basket default swaps

Definition 1.1.

A basket default swap is a credit derivative where the underlying is a

basket of reference entities.

If we want to hedge against losses in a portfolio of credits the natural way

is to buy insurance via one or several CDS or use a k-th to default swap.

A k-th to default swap is an insurance derivative which pays off the de-
fault leg when the kth entity defaults.

First to default swap

In the case of a portfolio of credits over n entities with exposures equal to
N1 , ..., Nn , recovery rates R1 , ..., Rn , and maturities equal to T, the first to
default swap would payout (1 − Ri )Ni in case entity i defaults first.

Formally, the stochastic value of the default leg is SVt (DL) =

R τ1:n rs ds
X · 1τ1:n ≤T · e− t where X = ni=1 1τ1:n =τi · (1 − Ri ) · Ni = (1 − Ri∗ ) · Ni
P

1
and τ1:n is the first order statistic among the default times τi , i = 1, n.
P
The stochastic value of the premium leg is SVt (PL) = c·N · ∆tm ·
− ttm rs ds
R
1τ1:n >tm e

Property 1.1.
Pn
max(sCDS
1 , sCDS
2 , ..., sCDS
n ) ≤ sF tD ≤ i=1 sCDS
i .
Proof :

If, by absurd, sF tD < max(sCDS1 , ..., sCDS

n ) ⇒ we could short any of the
CDS’s, and simultaneously go long on an FtD swap. When (If) a default
occurs, the potential loss (if the exact entity over whose CDS we have entered
defaults) is covered by the FtD but we remain with at least the net coupons
from the L/S strategy, whichPn areCDS positive.
F tD
If, by absurd, s > i=1 si we could go long on the n CDS’s and
short on the FtD. In any default case, we would be covered by the CDS and
still would remain with a positive net payoff from the long/short position.

Remark 1.1.

The first inequality becomes equality when the defaults are perfectly cor-
related, so one default would be followed by other defaults, meaning that
the FtD spread would be equal to the worst spread. The second inequal-
ity becomes equality when the defaults are uncorrelated so we should buy all
the CDS’s to be sure we have the same effect as over a first-to-default spread.

As in the single CDS case, in case of a k-th to default (KtD) swap

and in a particular FtD swap, we have the stochastic relation of sKtD =
P E[X ·1τ1:n ≤T ·Bt (τ1:n )] .
N tm ≥t∆tm ·S1:n,t (tm )·Bt (tm )

If the portfolio of credits is homogenous we have the following relation

holding: R T R T
(1−R) t Bt (u)fk:n,t (u)du (1−R) Bt (u)fk:n,t (u)du
sktD = P u RT t
. (H)
tm ≥t ∆tm Sk:n,t (tm )Bt (tm ) t Bt (u)Sk:n,t (u)du

In particular, we have the next:

2
Property 1.2.
If we have an homogenous credit portfolio over n entities (the same re-
covery rate R = R1 = ... = Rn and the same notional N ) whose default
times are represented by the independent random variables X1 , ..., PXn ∼
E(λ1 ), E(λ2 ), ..., E(λn ), then the first to default spread sF tD = (1−R) ni=1 λi
.
Proof :
Pn
We will use the fact that X1:n ∼ E( i=1 λi )

f1:n,t (u) = dP (Xdu

1:n <u)
= d(1−P (X
du
1:n >u))
= −d(P (X1 >u)P (Xdu2 >u)...P
Pn
(Xn >u))
=
1−e−(r+Pi=1 λi )(T −t)
Pn T
n+1
P n − i=1 λi
R Pn
(−1) ( i=1 λi )e ⇒ t Bt (u)f1:n,t (u)du = ... = r+ n i=1 λi
i=1 λi
(R1)
RT 1−e−(r+λ1 +λ2 +...+λn )(T −t)
t
S1:n,t (u)Bt (u)du = ... = r+λ1 +λ2 +. ..+λn
(R2)

From (R1), (R2) and remark H above, it results our statement.

k th - to default swaps

The k th to default swaps behave the same as a FtD swap, the default
leg is composed only of the entity defaulting at the time, so if we want to
hedge against several defaults we should buy either several CDS’s or several
KtD swaps. The cost usually is not the same. While a portfolio of CDS’s
offeres a targeted protection, a portfolio of kth to default swaps offers an
un-discriminatory protection against whoever defaults first.
However, we have the following no-arbitrage equality between the spreads.

Property 1.3.
For a portfolio of credits over n entities, sCDS
1 + sCDS
2 + ... + sCDS
n =
F tD StD LtD
s +s + ... + s .

Proof

To fully hedge the portfolio of n credits, we can either buy an entire port-
folio of CDS contracts or, buy a FtD swap, a StD swap, a TtD swap, etc.

3
The net effect is the same, so the net payments, should be the same. .

Let’s assume again we have the credit curves from table 1 given in the
previous paper (Calibration of PD).

Table 1: Credit spread curves

Maturity \Entity A B C
6M 130 bps 530 bps 190 bps
1Y 195 bps 490 bps 210 bps
3Y 240 bps 470 bps 250 bps
5Y 370 bps 400 bps 230 bps

Example 1.1.
An investor X (buy-and-hold type) has a portfolio of three bonds issued by
A,B and C, 3Y maturities all, notionals being NA = NB = NC = N = 1000$
and annual payments of coupons at rates cA = 4%, cB = 5%, cC = 6%. We
assume the entities A,B,C have recovery rates RA = RB = RC = 40%.

a. If insurance company I1 issues a 3Y FtD swap , spread s = 3% with

coupons paid annually, what would happen in the following scenarios?
1. No default occurs throughout the lifetime of the swap.
2. Entity B defaults after 6 months.
3. Entity C defaults after 1Y1M.

1. We pay 3% ×1000 = 30 $ each year on the FtD swap. The PnL to I1

is 90 $.
2. At the default time, X receives (1 − RB ) · NB = 600$ from insurer I1 while
he would pay the accrued premium of ∆t × s × N = 15$ to the insurer after
1Y (or a discounted value after 6M, discounted with the yield yB ), the payoff
to X being 600 - 15 = 585 $.
3. A full 30 $ is paid after 1Y and an accrued payment of 30/12 = 2.5 is
recorded after a month. In return X receives 600 so PnL = 600-32.5 = 568.5 $

b. If insurance company I2 issues a similar Second-to-default swap to

investor X, what would happen in the following scenarios?

4
1. No default occurs throughout the lifetime of the StD swap.
2. Entity B defaults after 6M but no other default occurs after that.
3. Entity C defaults after 1Y1M but no other default occurs after that.
4. Entity B defaults after 6M and entity C defaults after 2Y3M. (Added
scenario)
5. What would be the difference between having the following hedging port-
folios?
P1 = {FtD (from a),StD (b)}; P2 = {CDSB , CDSC }; P3 = {CDSA , CDSC }

1. Identical payments of 30 $ are made throughout the 3 years by X.

2. After the first year, X would pay an accrued premium of 5 $ for the
bond B (corresponding to the first 6M lifetime of the swap) and 3% × 1000 ×
2
3
= 20$ ⇒ 25$ the first year.

The second and third year I would pay 20$ each so the PnL for I2 =
25 + 20 + 20 = 65 $.

3. The first year X would pay 30 $, the second year I would pay 20 + the
1
1M accrued coupon corresponding to entity C (=10 × 12 ) = 0.833 $ therefore
20.8333 $ and the third, it would be 20 $ so the PnL to I2 = 70.8333 $.

4. In this case, X pays 25 $ the first year (as in case 1), 20 $ the second
but he would only pay the remaining accrued coupon from bond C and A
3
the third year, that is 20 × 12 = 5$ and in return he would receive 600 $ for
the defaulting bond C. The contract activates and its payoff for X would be
in that case 550 (= 600 - (25+20+5)) and for the insurer I2 the opposite.

5. If we look at the table of credit curves, portfolio P1 would cost annually

60 bps ×1000 = 60 $, portfolio P2 would cost sB (3) × 1000 + sC (3) × 1000 =
47 + 25 = 72$ while portfolio P3 would cost 49. So in the first scenario,
(that no defaults occur), the third portfolio would have the best PnL for the
insurance buyer X.
In the second scenario, the FtD from P1 would activate and also the CDS
for B from P2 would also. However portfolio P3 would not offer any protection
in this scenario.
In the third scenario, all portfolios offer protection, the least expensive is
again P3 .

5
 In the fourth scenario, P3 does not cover the default of B.

Conclusion:
In order to make a decision, one must compute the expected loss for each
covered portfolio and according to that choose the ”optimal” one. That is
reserved for another paper.

Example 1.2.

Let’s consider again the previous situation.

Q1: If X wants to be protected against the default of either of the three

entities what solutions can one propose? What would be the annual cost?

Q2: If an investor wants to be protected and can accept the default of

one entity but not of two entities what would be the annual cost?

Answers:

Q1: To the first question there are at least two choices:

1. buy a CDS on each bond, or
2. enter an FtD swap.

3. A riskier solution would be to hedge only the most likely to default,

that is to enter a CDS protecting the notional of bond B.

What happens in case of default of one of these bonds?

In case 1, the investor would lose the remaining coupons on the bond
that defaults but (s)he will be protected on the remaining exposure. (S)He
would pay (240bps+470bps+250bps) ·1000 = 960 bps ·1000 = 96$.
In case 2, the investor would lose the remaining coupons on the default-
ing bond, he recovers the notional as in case 1, the LGD from the CDS, but
he will have uncovered positions on the remaining bonds.
If the default times are considered independent, according to property
1.2 the investor would pay in this case the sum of the spreads (as in case 1)
= 96 $ per year.

6
 If the default times are not independent, which is most likely, the spread
F tD
s would be less so the FtD swap would be cheaper than buying CDS on
each bond.
In case 3, the investor could be in the same situation as in case 2 (no-
default or default for bond B), but he could as well lose on other uncovered
positions (A or C) and still pay protection on bond B. On the other hand,
he would pay less than in case 1 or 2, (according to Property 1.1) because
the coupon paid on a first to default swap, sF tD > max(sCDS
1 , sCDS
2 , sCDS
3 ).

Q2: In this case too there are several possibilities.

1. Buy an FtD and an StD swap. It is more costly, and according to the
investor’s risk profile, he would accept the default of a bond, but not of two.

2.R Buy only the StD swap. It would cost him(her) annually sStD =
(1−R) 03 f2:3 (t)B(t)dt
R3 , less than in the first case.
0 S2:3 (t)B(t)dt

In order to compute f2:3 (t), S2:3 (t) we must choose amongst the models
we have calibrated previously in paper 1, if we assume the default times are
independent, such as from exponential / piecewise exponential models.

If τ1 , τ2 , τ3 ∼ E(λ1 ), E(λ2 ), E(λ3 ) then S2:3 (t) = 1−[(e−λ1 t )+e−λ2 t +e−λ3 t −

2·((λ1 +λ2 )e−(λ1 +λ2 )t +(λ2 +λ3 )e−(λ2 +λ3 )t +(λ1 +λ3 )e−(λ1 +λ3 )t )+3e−(λ1 +λ2 +λ3 )t ]

If we choose the exponential model, calibrated on each entity as in the pre-

A (3) sB (3) sC (3)
vious article (PD calibration), having the parameters λ1 , λ2 , λ3 = s1−R , 1−R , 1−R =
0.024 0.047 0.025
0.6
, 0.6 , 0.6 , we obtain after some computations that the second to de-
fault spread is approximately 0.8%, much less than the First to Default Swap
spread. (Through direct integration or via Monte Carlo simulation of the de-
fault times).
That could be explained by the fact that, in case of one default, the StD
spread pays out 0 while an FtD spread pays out (1 − R) · 1000, and the fact
that the likelihoods of defaults are very spreaded.

3. The third choice would be to hedge against the two most likely bonds
to default (bonds B and C, that have the highest 3Y spread).