lip ton 2010
TRANSCRIPT
-
8/3/2019 Lip Ton 2010
1/145
CVA for CDS via SDM and Related IssuesAnalytical and Numerical Results
Alexander Lipton and Artur Sepp
Bank of America Merrill Lynch
March 2010
Lipton and A Sepp (Bank of America Merrill Lynch)
Credit Modeling 02/03 1 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
2/145
Plan
A brief history of credit modeling.
Lipton and A Sepp (Bank of America Merrill Lynch)
Credit Modeling 02/03 2 / 74
http://find/ -
8/3/2019 Lip Ton 2010
3/145
Plan
A brief history of credit modeling.
How to model CDSs?
Lipton and A Sepp (Bank of America Merrill Lynch)
Credit Modeling 02/03 2 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
4/145
Plan
A brief history of credit modeling.
How to model CDSs?
Reduced, structural, and hybrid single-name models.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 2 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
5/145
Plan
A brief history of credit modeling.
How to model CDSs?
Reduced, structural, and hybrid single-name models.Examples.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 2 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
6/145
Plan
A brief history of credit modeling.
How to model CDSs?
Reduced, structural, and hybrid single-name models.Examples.
Multi-name structural models.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 2 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
7/145
Plan
A brief history of credit modeling.
How to model CDSs?
Reduced, structural, and hybrid single-name models.Examples.
Multi-name structural models.
Examples.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 2 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
8/145
Acknowledgements
We am also grateful to other Merrill Lynch colleagues, especially S.
Inglis, A. Rennie, D. Shelton, I. Savescu, and members of the CreditAnalytics Team and Structured Credit Team for all their help.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 3 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
9/145
References
Selected references:
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 4 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
10/145
References
Selected references:
Reduced form model: Jarrow & Turnbull (2000), Hughston &Turnbull (2001), Lando (1998), Madan & Unal (1994).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 4 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
11/145
References
Selected references:
Reduced form model: Jarrow & Turnbull (2000), Hughston &Turnbull (2001), Lando (1998), Madan & Unal (1994).
Structural model: Black & Cox (1976), Hilberink & Rogers (2002),Hyer et al (1998), Kiesel & Scherer (2007), Leland (1994), Longsta& Schwartz (1995), Lipton (2002), Merton (1974), Sepp (2004,
2006), Zhou (2001a), Zhou (2001a).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 4 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
12/145
References
Selected references:
Reduced form model: Jarrow & Turnbull (2000), Hughston &Turnbull (2001), Lando (1998), Madan & Unal (1994).
Structural model: Black & Cox (1976), Hilberink & Rogers (2002),Hyer et al (1998), Kiesel & Scherer (2007), Leland (1994), Longsta& Schwartz (1995), Lipton (2002), Merton (1974), Sepp (2004,
2006), Zhou (2001a), Zhou (2001a).
Counterparty risk: Blanchet & Patras (2008), Brigo & Chourdakis(2008), Crepey, Jeanblanc & Zargari (2009), Leung & Kwok (2005),Haworth, Reisisnger & Shaw (2006), Hull & White (2000), (2001),
Misisrpashayev (2008), Turnbull (2005), Valuzis (2008).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 4 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
13/145
References
Selected references:
Reduced form model: Jarrow & Turnbull (2000), Hughston &Turnbull (2001), Lando (1998), Madan & Unal (1994).
Structural model: Black & Cox (1976), Hilberink & Rogers (2002),Hyer et al (1998), Kiesel & Scherer (2007), Leland (1994), Longsta& Schwartz (1995), Lipton (2002), Merton (1974), Sepp (2004,
2006), Zhou (2001a), Zhou (2001a).
Counterparty risk: Blanchet & Patras (2008), Brigo & Chourdakis(2008), Crepey, Jeanblanc & Zargari (2009), Leung & Kwok (2005),Haworth, Reisisnger & Shaw (2006), Hull & White (2000), (2001),
Misisrpashayev (2008), Turnbull (2005), Valuzis (2008).Numerical methods: Andersen & Andreasen (2000), Boyarchenko &Levendorsky (2000), Cont & Tankov (2004), Cont & Voltchkova(2005), Clift & Forsyth (2008), dHalluin, Forsyth & Vetzal (2005),Feng & Linetsky (2008), Lipton (2003).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 4 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
14/145
Moodys number of defaults gure
All rated, Source Moody's
0
50
100
150
200
250
1920
1924
1928
1932
1936
1940
1944
1948
1952
1956
1960
1964
1968
1972
1976
1980
1984
1988
1992
1996
2000
2004
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 5 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
15/145
Moodys table of recoveries
Source: Moody'sYear Sr. Secured Sr. Secured Sr. Unsecured Sr. SubordinaSubordinated Jr. SubordAll
Bank Loans Bonds Bonds Bonds Bonds Bonds Bonds
1982 NA $72.50 $35.79 $48.09 $29.99 NA $35.571983 NA $40.00 $52.72 $43.50 $40.54 NA $43.641984 NA NA $49.41 $67.88 $44.26 NA $45.491985 NA $83.63 $60.16 $30.88 $39.42 $48.50 $43.661986 NA $59.22 $52.60 $50.16 $42.58 NA $48.381987 NA $71.00 $62.73 $44.81 $46.89 NA $50.481988 NA $55.40 $45.24 $33.41 $33.77 $36.50 $38.981989 NA $46.54 $43.81 $34.57 $26.36 $16.85 $32.311990 $75.25 $33.81 $37.01 $25.64 $19.09 $10.70 $25.50
1991 $74.67 $48.39 $36.66 $41.82 $24.42 $7.79 $35.531992 $61.13 $62.05 $49.19 $49.40 $38.04 $13.50 $45.891993 $53.40 NA $37.13 $51.91 $44.15 NA $43.081994 $67.59 $69.25 $53.73 $29.61 $38.23 NA $45.571995 $75.44 $62.02 $47.60 $34.30 $41.54 NA $43.281996 $88.23 $47.58 $62.75 $43.75 $22.60 NA $41.541997 $78.75 $75.50 $56.10 $44.73 $35.96 $30.58 $49.391998 $51.40 $48.14 $41.63 $44.99 $18.19 $62.00 $39.651999 $75.82 $43.00 $38.04 $28.01 $35.64 NA $34.332000 $68.32 $39.23 $23.81 $20.75 $31.86 $15.50 $25.18
2001 $66.16 $37.98 $21.45 $19.82 $15.94 $47.00 $22.212002 $58.80 $48.37 $29.69 $23.21 $24.51 NA $30.182003 $73.43 $63.46 $41.87 $37.27 $12.31 NA $40.692004 $87.74 $73.25 $54.25 $46.54 $94.00 NA $59.122005 $82.07 $71.93 $54.88 $26.06 $51.25 NA $55.972006 $76.02 $74.63 $55.02 $41.41 $56.11 NA $55.022007 $67.74 $80.54 $51.02 $54.47 NA NA $53.53
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 6 / 74
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
16/145
CDS gure
0
50
100
150
200
250
300
350
400
29/3/07 28/5/07 27/7/07 25/9/07 24/11/07 23/1/08 23/3/08
IBM 5Y GE 5Y MER 5Y DB 5Y DCX 5Y SIEM 5Y
Figure: Time series of 5Y CDS spreads (in bps) for 6 typical companies (IBM, GE,Merrill Lynch, Deutsche Bank, Daimler, and Siemens). Source: Merrill Lynch.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 7 / 74
R d
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
17/145
Recent trends
The volume of CDSs grew by a factor of 100 between 2001 and 2007.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 8 / 74
R d
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
18/145
Recent trends
The volume of CDSs grew by a factor of 100 between 2001 and 2007.
Lately, CDS dealers reduced their market by 38% in one year.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 8 / 74
R d
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
19/145
Recent trends
The volume of CDSs grew by a factor of 100 between 2001 and 2007.
Lately, CDS dealers reduced their market by 38% in one year.
According to the most recent ISDA survery published on April 22nd,the notional value of CDSs outstanding decreased to $38.6 trillion asof Dec 31st 2008 from $54.6 trillion mid-year and $62.2 trillion as ofDec 31st 2007.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 8 / 74
CDS d li
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
20/145
CDS modeling
First, we need to explain how to price CDSs.
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 9 / 74
CDS d li
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
21/145
CDS modeling
First, we need to explain how to price CDSs.
Then we need to extend our theory to cover indices, tranches,baskets, etc.
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 9 / 74
CDS modeling
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
22/145
CDS modeling
First, we need to explain how to price CDSs.
Then we need to extend our theory to cover indices, tranches,baskets, etc.
Three complementary approaches to pricing CDSs:
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 9 / 74
CDS modeling
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
23/145
CDS modeling
First, we need to explain how to price CDSs.
Then we need to extend our theory to cover indices, tranches,baskets, etc.
Three complementary approaches to pricing CDSs:reduced form,
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 9 / 74
CDS modeling
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
24/145
CDS modeling
First, we need to explain how to price CDSs.
Then we need to extend our theory to cover indices, tranches,baskets, etc.
Three complementary approaches to pricing CDSs:reduced form,structural,
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 9 / 74
CDS modeling
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
25/145
CDS modeling
First, we need to explain how to price CDSs.
Then we need to extend our theory to cover indices, tranches,baskets, etc.
Three complementary approaches to pricing CDSs:reduced form,structural,hybrid
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 9 / 74
CDS modeling
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
26/145
CDS modeling
First, we need to explain how to price CDSs.
Then we need to extend our theory to cover indices, tranches,baskets, etc.
Three complementary approaches to pricing CDSs:reduced form,structural,hybrid
All of them have pros and cons. Reduced form and structural
approaches dominate.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 9 / 74
Simple non-risky model
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
27/145
Simple non-risky model
First, we consider non-risky bonds and introduce the short interestrate r(t) which is driven by the SDE:
dr(t) = f1 (t, r (t)) dt+ g1 (t, r(t)) dW1 (t) r(0) = r0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 10 / 74
Simple non-risky model
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
28/145
Simple non risky model
First, we consider non-risky bonds and introduce the short interestrate r(t) which is driven by the SDE:
dr(t) = f1 (t, r (t)) dt+ g1 (t, r(t)) dW1 (t) r(0) = r0
Let D(t, T) be time t price of the non-risky zero-coupon bondmaturing at time T. This price can be written as follows:
D(
t, T) =
V1 (
t, r(
t))
D(0
, T) =
V1 (0
, r0)
where V1 (t, r) is the solution of the following backward Kolmogorovproblem
V1,t +12 g
21 V1,rr + f1V1,r rV1 = 0 V1 (T, r) = 1
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 10 / 74
Simple non-risky model
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
29/145
Simple non risky model
First, we consider non-risky bonds and introduce the short interestrate r(t) which is driven by the SDE:
dr(t) = f1 (t, r (t)) dt+ g1 (t, r(t)) dW1 (t) r(0) = r0
Let D(t, T) be time t price of the non-risky zero-coupon bondmaturing at time T. This price can be written as follows:
D(
t, T) =
V1 (
t, r(
t))
D(
0, T) =
V1 (
0, r0)
where V1 (t, r) is the solution of the following backward Kolmogorovproblem
V1,t +12 g
21 V1,rr + f1V1,r rV1 = 0 V1 (T, r) = 1
Functions f1, g1 are calibrated to the market to match the yield curveand some swaption volatilities. Alternatively,
D(t, T) = Et
exp
ZT
tr
t0
dt0
= exp
ZT
tr
t, t0
dt0
where r is the forward rate.Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 10 / 74
Reduced form model formulation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
30/145
Reduced form model formulation
The name defaults at the rst time a Cox process jumps from 0 to 1.SDE for the default intensity (hazard rate) X (t) is:
dX (t) = f(t, X (t)) dt+ g (t, X (t)) dW (t) + JdN(t) , X (0) = X0
where W (t) is a standard Wiener processes, N(t) is a Poissonprocess with intensity (t), and J is a positive jump distribution;W, N, J are mutually independent.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 11 / 74
Reduced form model formulation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
31/145
Reduced form model formulation
The name defaults at the rst time a Cox process jumps from 0 to 1.SDE for the default intensity (hazard rate) X (t) is:
dX (t) = f(t, X (t)) dt+ g (t, X (t)) dW (t) + JdN(t) , X (0) = X0
where W (t) is a standard Wiener processes, N(t) is a Poissonprocess with intensity (t), and J is a positive jump distribution;W, N, J are mutually independent.
We impose the following constraints
f(t, 0) 0, f(t,) < 0, g (t, 0) = 0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 11 / 74
Reduced form model formulation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
32/145
The name defaults at the rst time a Cox process jumps from 0 to 1.SDE for the default intensity (hazard rate) X (t) is:
dX (t) = f(t, X (t)) dt+ g (t, X (t)) dW (t) + JdN(t) , X (0) = X0
where W (t) is a standard Wiener processes, N(t) is a Poissonprocess with intensity (t), and J is a positive jump distribution;W, N, J are mutually independent.
We impose the following constraints
f(t, 0) 0, f(t,) < 0, g (t, 0) = 0For analytical convenience (rather than for deeper reasons) it is
customary to assume that X is governed by the square-root stochasticdierential equation (SDE):
dX (t) = ( (t) X (t)) dt+ q
X (t)dW (t) + JdN(t) , X (0) = X0
with exponential (or hyper-exponential) jump distribution.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 11 / 74
Survival probability
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
33/145
p y
For practical purposes it is more convenient to consider discrete jumpdistributions with jump values Jm > 0, 1
m
M, occurring with
probabilities m > 0; such distributions are more exible thanparametric ones because they allow one to place jumps where they areneeded.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 12 / 74
Survival probability
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
34/145
p y
For practical purposes it is more convenient to consider discrete jumpdistributions with jump values Jm > 0, 1
m
M, occurring with
probabilities m > 0; such distributions are more exible thanparametric ones because they allow one to place jumps where they areneeded.In this framework, the survival probability of the name from time 0 totime T has the form
q(0, T) = E0n
eRT
0 X(t0)dt0
o= E0
neY(T)
o
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 12 / 74
Survival probability
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
35/145
p y
For practical purposes it is more convenient to consider discrete jumpdistributions with jump values Jm > 0, 1
m
M, occurring with
probabilities m > 0; such distributions are more exible thanparametric ones because they allow one to place jumps where they areneeded.In this framework, the survival probability of the name from time 0 totime T has the form
q(0, T) = E0n
eRT
0 X(t0)dt0
o= E0
neY(T)
oHere Y (t) is governed by the following degenerate SDE:
dY (t) = X (t) dt, Y (0) = 0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 12 / 74
Survival probability
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
36/145
For practical purposes it is more convenient to consider discrete jumpdistributions with jump values Jm > 0, 1
m
M, occurring with
probabilities m > 0; such distributions are more exible thanparametric ones because they allow one to place jumps where they areneeded.In this framework, the survival probability of the name from time 0 totime T has the form
q(0, T) = E0n
eRT
0 X(t0)dt0
o= E0
neY(T)
oHere Y (t) is governed by the following degenerate SDE:
dY (t) = X (t) dt, Y (0) = 0
More generally, the survival probability from time t to time Tconditional on no default before time t has the form
q( t, TjX (t) , Y (t)) = eY(t)I(>t)Etn
eY(T)
X (t) , Y (t)o
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 12 / 74
Calculation of expectations
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
37/145
Expectations of the form Et
neY(T)
X (t) , Y (t)
o, can be
computed by solving the following augmented partial dierentialequation (PDE)
(t + L) V (t, T, X, Y) + XVY (t, T, X, Y) = 0
V (T, T, X, Y) = eY
where
LV ( (t) X) VX + 12 2XVXX + m
m [V (X + Jm ) V (X)]
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 13 / 74
Calculation of expectations
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
38/145
Expectations of the form Et
neY(T)
X (t) , Y (t)
o, can be
computed by solving the following augmented partial dierentialequation (PDE)
(t + L) V (t, T, X, Y) + XVY (t, T, X, Y) = 0
V (T, T, X, Y) = eY
where
LV ( (t) X) VX + 12 2XVXX + m
m [V (X + Jm ) V (X)]
Specically, the following relation holds
Et
neY(T)
X (t) , Y (t)o
= V (t, T, X (t) , Y (t))
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 13 / 74
Ane ansatz
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
39/145
The corresponding solution can be written in the so-called ane form:
V (t, T, X, Y) = ea(t,T,)+b(t,T,)XY
where a, b are functions of time governed by the following system ofordinary dierential equations (ODEs):
8
-
8/3/2019 Lip Ton 2010
40/145
The corresponding solution can be written in the so-called ane form:
V (t, T, X, Y) = ea(t,T,)+b(t,T,)XY
where a, b are functions of time governed by the following system ofordinary dierential equations (ODEs):
8
-
8/3/2019 Lip Ton 2010
41/145
The value U of a credit default swap (CDS) paying an up-frontamount and a coupon s in exchange for receiving 1
R (where R is
the default recovery) on default as follows:
U = + V (0, X0)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 15 / 74
CDS valuation 1
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
42/145
The value U of a credit default swap (CDS) paying an up-frontamount and a coupon s in exchange for receiving 1
R (where R is
the default recovery) on default as follows:
U = + V (0, X0)Here V (t, X) solves the following pricing problem
(t + L) V (t, X) (r + X) V (t, X) = s (1 R) XV (T, X) = 0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 15 / 74
CDS valuation 1
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
43/145
The value U of a credit default swap (CDS) paying an up-frontamount and a coupon s in exchange for receiving 1
R (where R is
the default recovery) on default as follows:
U = + V (0, X0)Here V (t, X) solves the following pricing problem
(t + L) V (t, X) (r + X) V (t, X) = s (1 R) XV (T, X) = 0
Using Duhamels principle, we obtain the following expression for V:
V (t, X) = sZT
t D
t, t0
ea(t,t
0,1)+b(t,t
0,1)X
dt0
(1 R)ZT
tD
t, t0
dh
ea(t,t0,1)+b(t,t0,1)X
iwhere D(t, t0) is the discount factor between two times t, t0.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 15 / 74
CDS valuation 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
44/145
Accordingly,
U = sZT
0 D
0, t0
1 p0, t0 dt0 + (1 R) ZT
0 D
0, t0
dp
For a given up-front payment , we can represent the correspondingpar spread s (i.e. the spread which makes the value of thecorresponding CDS zero) as follows:
s(T) = + (1 R)R
T0 D(0, t0) dp(0, t0)RT
0 D(0, t0) (1 p(0, t0)) dt0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 16 / 74
CDS valuation 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
45/145
Accordingly,
U = sZT
0 D
0, t0
1 p0, t0 dt0 + (1 R) ZT
0 D
0, t0
dp
For a given up-front payment , we can represent the correspondingpar spread s (i.e. the spread which makes the value of thecorresponding CDS zero) as follows:
s(T) = + (1 R)R
T0 D(0, t0) dp(0, t0)RT0 D(0, t
0) (1 p(0, t0)) dt0Conversely, for a given spread we can represent the par up-frontpayment in the form
= s(T)ZT
0D
0, t0
1 p0, t0 dt0+ (1 R) ZT0
D
0, t0
dp
0, t0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 16 / 74
CDS valuation 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
46/145
Accordingly,
U = sZT
0 D
0, t0
1 p0, t0 dt0 + (1 R) ZT
0 D
0, t0
dp
For a given up-front payment , we can represent the correspondingpar spread s (i.e. the spread which makes the value of thecorresponding CDS zero) as follows:
s(T) = + (1 R)R
T0 D(0, t0) dp(0, t0)RT0 D(0, t
0) (1 p(0, t0)) dt0Conversely, for a given spread we can represent the par up-frontpayment in the form
= s(T)Z
T
0D
0, t0
1 p0, t0 dt0+ (1 R) ZT0
D
0, t0
dp
0, t0
In these formulas we implicitly assumed that the corresponding CDSis fully collateralized, so that in the event of default 1 R is readilyavailable.Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 16 / 74
Two name correlation 1
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
47/145
It is very tempting to extend the above framework to cover severalcorrelated names. For example, consider two credits, A, B and assume
for simplicity that their default intensities coincide,
XA (t) = XB (t) = X (t)
and both names have the same recovery RA = RB = R.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 17 / 74
Two name correlation 1
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
48/145
It is very tempting to extend the above framework to cover severalcorrelated names. For example, consider two credits, A, B and assume
for simplicity that their default intensities coincide,
XA (t) = XB (t) = X (t)
and both names have the same recovery RA = RB = R.
For a given maturity T, the default event correlation is dened asfollows
(0, T) =pAB (0, T) pA (0, T) pB (0, T)
ppA (0, T) (1 pA (0, T)) pB (0, T) (1 pB (0, T))
where A, B are the default times, andpA (0, T) = P(A T) , pB (0, T) = P(B T)
pAB (0, T) = P(A T, B T)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 17 / 74
Two name correlation 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
49/145
It is clear that
pA (0, T) = pB (0, T) = p(0, T) = 1
ea(0,T,1)+b(0,T,1)X0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 18 / 74
Two name correlation 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
50/145
It is clear that
pA (0, T) = pB (0, T) = p(0, T) = 1
ea(0,T,1)+b(0,T,1)X0
Simple calculation yields
pAB (0, T) = E0n
eRT
0 (XA (t0)+XB(t0))dt0
o+pA (0, T) +pB (0, T)1
=E
0ne2 R
T
0 X(t0)dt0o
+ 2p
(0, T
) 1
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 18 / 74
Two name correlation 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
51/145
It is clear that
pA (0, T) = pB (0, T) = p(0, T) = 1
ea(0,T,1)+b(0,T,1)X0
Simple calculation yields
pAB (0, T) = E0n
eRT
0 (XA (t0)+XB(t0))dt0
o+pA (0, T) +pB (0, T)1
=E
0ne2 R
T
0 X(t0)dt0o
+ 2p
(0, T
) 1Thus
(0, T) =ea(0,T,2)+b(0,T,2)X0 e2a(0,T,1)+2b(0,T,1)X0
1 ea(0,T,1)+b(0,T,1)X0
ea(0,T,1)+b(0,T,1)X0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 18 / 74
Two name correlation 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
52/145
It is clear that
pA (0, T) = pB (0, T) = p(0, T) = 1
ea(0,T,1)+b(0,T,1)X0
Simple calculation yields
pAB (0, T) = E0n
eRT
0 (XA (t0)+XB(t0))dt0
o+pA (0, T) +pB (0, T)1
=E
0ne2 R
T
0 X(t0)dt0o
+ 2p
(0, T
) 1Thus
(0, T) =ea(0,T,2)+b(0,T,2)X0 e2a(0,T,1)+2b(0,T,1)X0
1 ea(0,T,1)+b(0,T,1)X0
ea(0,T,1)+b(0,T,1)X0
In the absence of jumps, the corresponding event correlation is verylow. If large positive jumps are added (while overall survivalprobability is preserved), then correlation can increase all the way toone. Assuming that T = 5y, = 0.5, = 7%, and J = 5.0, weillustrate this observation below.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 18 / 74
Event correlation Figure
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
53/145
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
0.00% 0.50% 1.00% 1.50% 2.00% 2.50% 3.00%
lambda
0.00%
0.50%
1.00%
1.50%
2.00%
2.50%
3.00%
3.50%
corr (left axis) theta (right axis)
Figure: Correlation and mean-reversion level = X0 as functions of jumpintensity . Other parameters are as follows: T = 5y, = 0.5, = 7%, andJ
= 5.0.Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 19 / 74
FTD-STD
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
54/145
In the two-name portfolio, we can dene two types of CDSs which
depend on the correlation: (A) the rst-to-default swap (FTD); (B)the second-to-default swap (STD). The corresponding par spreads(assuming that there are no up-front payments) are
s1 (T) =
(1
R) RT0 D(0, t0) dh1 e
a(0,t0,2)+b(0,t0,2)X0iRT0 D(0, t
0) ea(0,t0,2)+b(0,t0,2)X0 dt0
s2 (T) =(1 R)
RT
0D(0, t0) d
h1
2ea(t
0,1)+b(t0,1)X0 ea(t0,2)+b(t0,2)X
RT0 D(0, t0)
2e
a(
t0,
1)+b
(t0,
1)X
0 ea
(t0,
2)+b
(t0,
2)X
0
dt0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 20 / 74
FTD-STD
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
55/145
In the two-name portfolio, we can dene two types of CDSs which
depend on the correlation: (A) the rst-to-default swap (FTD); (B)the second-to-default swap (STD). The corresponding par spreads(assuming that there are no up-front payments) are
s1 (T) =
(1
R) RT0 D(0, t0) dh1 e
a(0,t0,2)+b(0,t0,2)X0iRT0 D(0, t
0) ea(0,t0,2)+b(0,t0,2)X0 dt0
s2 (T) =(1 R)
RT
0D(0, t0) d
h1
2ea(t
0,1)+b(t0,1)X0 ea(t0,2)+b(t0,2)X
RT0 D(0, t0)
2e
a(
t0,
1)+b
(t0,
1)X
0 ea
(t0,
2)+b
(t0,
2)X
0
dt0It is clear that the relative values of s1, s2 very strongly depend onwhether or not jumps are present in the model.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 20 / 74
FTD-STD Figure
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
56/145
0.00%
0.50%
1.00%
1.50%
2.00%
2.50%
3.00%
3.50%
4.00%
0.00% 0.50% 1.00% 1.50% 2.00% 2.50% 3.00%
lambda
s_1 s_2 s
Figure: FTD spread s1, STD spread s2, and single name CDS spread s asfunctions of jump intensity . Other parameters are the same as in Fig.1. It isclear that jumps are necessary to have s
1and s
2of similar magnitudes.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 21 / 74
Counter-party eects
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
57/145
An important application of the above model is to the evaluation ofcounter-party eects on fair CDS spreads. Let us assume that name
A has written a CDS on reference name B. It is clear that the pricingproblem for the value of the uncollateralized CDS V can be written asfollows
LV (t, X) (r + 2X) V (t, X)= s
(1
R) X
(RV+ (t, X) + V
(t, X)) X
where V is the value of a fully collateralized CDS on name B.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 22 / 74
Counter-party eects
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
58/145
An important application of the above model is to the evaluation ofcounter-party eects on fair CDS spreads. Let us assume that name
A has written a CDS on reference name B. It is clear that the pricingproblem for the value of the uncollateralized CDS V can be written asfollows
LV (t, X) (r + 2X) V (t, X)= s
(1
R) X
(RV+ (t, X) + V
(t, X)) X
where V is the value of a fully collateralized CDS on name B.It is clear that the discount rate is increased from r + X to r + 2X,since there are two cases when the uncollateralized CDS can beterminated due to default: when the reference name B defaults; when
the issuer A defaults.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 22 / 74
Counter-party eects
A i li i f h b d l i h l i f
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
59/145
An important application of the above model is to the evaluation ofcounter-party eects on fair CDS spreads. Let us assume that name
A has written a CDS on reference name B. It is clear that the pricingproblem for the value of the uncollateralized CDS V can be written asfollows
LV (t, X) (r + 2X) V (t, X)= s
(1
R) X
(RV+ (t, X) + V
(t, X)) X
where V is the value of a fully collateralized CDS on name B.It is clear that the discount rate is increased from r + X to r + 2X,since there are two cases when the uncollateralized CDS can beterminated due to default: when the reference name B defaults; when
the issuer A defaults.Although this equation is no longer analytically solvable, it can besolved numerically via, say, an appropriate modication of theclassical Crank-Nicholson method. It turns out that in the presence ofjumps the value of the fair par spread goes down dramatically.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 22 / 74
Counter-party Figure
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
60/145
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
0.00% 0.50% 1.00% 1.50% 2.00% 2.50% 3.00%
lambda
Fraction of Default Leg Value
Figure: Reduction in default leg value as function of jump intensity .Mean-reversion level = X0 is chosen in order to preserve survival probability.Other parameters are as follows: T = 5y, = 0.5, = 7%, and J = 5.0.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 23 / 74
Typical structural model without jumps formulation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
61/145
A typical structural model without jumps for the evolution of thelog-value of the rm has the form:
dx = dt+ dW (t) , x(0) = x0
x(t) = ln
v(t)B(t)
= ln
v (t)
B(0) exp ((r d) t)
= 2
2
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 24 / 74
Typical structural model without jumps formulation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
62/145
A typical structural model without jumps for the evolution of thelog-value of the rm has the form:
dx = dt+ dW (t) , x(0) = x0
x(t) = ln
v(t)B(t)
= ln
v (t)
B(0) exp ((r d) t)
= 2
2
This value is governed by a Wiener process. The rm defaults if thevalue x(t) crosses zero barrier b(t) = 0.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 24 / 74
Survival probability
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
63/145
Assuming that all the relevant parameters are constant, thecorresponding formula for Q(t, T) is
Q(t, T) = I (0, t)
N
x(t) 12 2p
2
! ex(t)N
x(t) 12 2p
2
!!
where N is the cumulative normal distribution. It is easy to verify that
Q(0, 0) = 1, QT (0, 0) = 0
The latter fact is rather disconcerting, since it implies that the parshort term CDS spread vanishes when T ! 0.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 25 / 74
Typical structural model with jumps formulation
A t ical st ct al odel ith j s fo the e ol tio of the
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
64/145
A typical structural model with jumps for the evolution of thelog-value of the rm has the form:
dx = dt+ dW (t) +j+dN+ +jdN, x(0) = x0
x = ln v
B
= 2
2 +
+ 1 +
+ 1
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 26 / 74
Typical structural model with jumps formulation
A typical structural model with jumps for the evolution of the
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
65/145
A typical structural model with jumps for the evolution of thelog-value of the rm has the form:
dx = dt+ dW (t) +j+dN+ +jdN, x(0) = x0
x = ln v
B
= 2
2 +
+ 1 +
+ 1This value is governed by a combination of a Wiener process and aPoisson process with exponentially distributed jumps. The rmdefaults if the value x(t) crosses zero barrier b(t) = 0. It was realized
early on that without jumps (or/and curvilinear or uncertain barriers)it is impossible to explain the short end of the CDS curve within thestructural framework. When all the relevant parameters are constant,the problem can be solved analytically via the Laplace transform.However, in general this approach does not work.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 26 / 74
Unbounded PDF 1
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
66/145
When the barrier is absent, can be found via a simple recursion
(knowing an analytical solution is useful for benchmarking purposes).Vis-a-vis the Gaussian distribution, has fat tails and a narrow peak.Let
=
p
t
= (x t) /pt0,0 (x) = n () /
pt
1,0 (x) = +P (,+)
0,
1 (x
) = P
(,)
where P (a, b) = exp
ab+ b2/2
N (a + b)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 27 / 74
Unbounded PDF 2
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
67/145
Then:
k,0 (x) =+ ((+ +)k1,0 + +k2,0)
k 10,l (x) =
(( )0,l1 + 0,l2)l
1
k,l (x) = +k1,l + k,l1
+ +
wk,l =e(++)t (+t)k (t)l
k!l!
(x) =
k=0,l=0
wk,lk,l (x)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 28 / 74
No barrier PDF gure
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
68/145
sigma=0.2, lambda=0.1, alpha=1.0, mu=0.0
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
-10.00 -8.00 -6.00 -4.00 -2.00 0.00 2.00 4.00
X
PDF
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 29 / 74
Fokker-Planck equation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
69/145
We can solve the barrier problem by using the forward Fokker-Planckequation for the t.p.d.f. and putting probabilities below the barrier tozero. This equation has the form:
L
t
12
2xx + x +
Z0
(t, x +j) ejdj = 0
(0, x) = (x ) (t, x) = 0 if < 0
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 30 / 74
Fokker-Planck equation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
70/145
We can solve the barrier problem by using the forward Fokker-Planckequation for the t.p.d.f. and putting probabilities below the barrier tozero. This equation has the form:
L
t
12
2xx + x +
Z0
(t, x +j) ejdj = 0
(0, x) = (x ) (t, x) = 0 if < 0
When parameters are time-independent, we can nd via theLaplace transform.
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 30 / 74
Laplace transform
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
71/145
The Laplace transform
1
22xxx ( + p) +
Z0 (p, x +j) ejdj = (x )
(p, 0) = 0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 31 / 74
Laplace transform
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
72/145
The Laplace transform
1
22xxx ( + p) +
Z0 (p, x +j) ejdj = (x )
(p, 0) = 0
Forward characteristic equation:
1
222 ( + p)
= 0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 31 / 74
Laplace transform
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
73/145
The Laplace transform
1
22xxx ( + p) +
Z0 (p, x +j) ejdj = (x )
(p, 0) = 0
Forward characteristic equation:
1
222 ( + p)
= 0
This equation has three roots of which two are positive and onenegative. We denote them as i.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 31 / 74
Greens function construction 1
Hence the overall solution has the form:
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
74/145
(p, x) = C3e3 (x); x
D1e
1(x
)
+ D2e
2(x
)
+ D3e
3(x
)
; x
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 32 / 74
Greens function construction 1
Hence the overall solution has the form:
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
75/145
(p, x) = C3e3 (x); x
D1e
1(x
)
+ D2e
2(x
)
+ D3e
3(x
)
; x Matching conditions
D1 + D2 + (D3 C3) = 0
1D
1+
2D
2+
3(D
3 C
3) =
2
2D1e
1 + D2e2 + D3e
3 = 0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 32 / 74
Greens function construction 1
Hence the overall solution has the form:
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
76/145
(p, x) = C3e3 (x); x
D1e
1(x
)
+ D2e
2(x
)
+ D3e
3(x
)
; x Matching conditions
D1 + D2 + (D3 C3) = 0
1D
1+
2D
2+
3(D
3 C
3) =
2
2D1e
1 + D2e2 + D3e
3 = 0
One more condition is obtained from the following observations:
L ei(x)H(x ) =
+ ie(x)H(
x)
L
ei(x)H( x)
= + i
e(x)H( x)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 32 / 74
Greens function construction 1
Hence the overall solution has the form:
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
77/145
(p, x) = C3e3 (x); x
D1e
1
(x
)
+ D2e
2
(x
)
+ D3e
3
(x
)
; x Matching conditions
D1 + D2 + (D3 C3) = 0
1D
1+
2D
2+
3(D
3 C
3) =
2
2D1e
1 + D2e2 + D3e
3 = 0
One more condition is obtained from the following observations:
L ei(x)H(x ) =
+ ie(x)H(
x)
L
ei(x)H( x)
= + i
e(x)H( x)
Here L is the corresponding dierential operator and H(.) is theHeaviside function.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 32 / 74
Greens function construction 2
The corresponding prole has the form
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
78/145
p g p
(p, x) =
C3e3 (x
)
; x D1e
1
e1 x e3 x + D2e2 e2 x e3x ; x
D1 =
2
2
( + 1)
(1 2) (1 3)D2 = 2
2( + 2)
(2 1) (2 3)
D3 =2
2" e
(13) ( + 1)
(1 2) (1 3)+
e(23 ) ( + 2)
(2 1) (2 3)#C3 = D1 + D2 + D3
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 33 / 74
Stehfest algorithm
Inverse Laplace transform yields (t, x). We use Stehfest algorithm
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
79/145
to evaluate Q:
(t, x) = pN
k=1
(1)k StNk (kp, x)
p =ln 2
t
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 34 / 74
Stehfest algorithm
Inverse Laplace transform yields (t, x). We use Stehfest algorithmQ
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
80/145
to evaluate Q:
(t, x) = pN
k=1
(1)k StNk (kp, x)
p =ln 2
t
Coecients StNk are very sti. We typically choose N = 20. For smallt inversion can be numerically unstable unless computation is carriedwith many signicant digits.
-15
-10
-5
0
5
10
15
20
25
30
35
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 34 / 74
Hitting time density
The rst hitting time density f(t) has the form
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
81/145
( )
f(t) = 2
2x (t, 0)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 35 / 74
Hitting time density
The rst hitting time density f(t) has the form
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
82/145
f(t) = 2
2x (t, 0)
Its Laplace transform is given by
f(p) =
e2 ( + 2)
e1 ( + 1)
2 1
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 35 / 74
Hitting time density
The rst hitting time density f(t) has the form
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
83/145
f(t) = 2
2 x (t, 0)
Its Laplace transform is given by
f(p) =
e2 ( + 2)
e1 ( + 1)
2 1For the regular diusion we obtain
f(p) = e2
f(t) = 2
2x (t, 0) =
e(+t)2/22 t
p22t3
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 35 / 74
Hitting time density gure
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
84/145
Hitting time density
0.000
0.002
0.004
0.006
0.0080.010
0.012
0.014
0.016
0.018
0.020
0.00 5.00 10.00 15.00 20.00 25.00
Time
Prob
Jump-Diffusion Inversion Regular Diffusion Inversion Regular Diffusion Analytical
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 36 / 74
Greens function expansion
Perturbative calculation of Greens function in 1D. For brevity weassume that all the parameters are time independent
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
85/145
assume that all the parameters are time independent.Time-dependent case can be solved in a similar way. To start with,we make the following transform
(t, x) = exp
2
22+
t+
2(x )
(t, x)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 37 / 74
Greens function expansion
Perturbative calculation of Greens function in 1D. For brevity weassume that all the parameters are time independent
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
86/145
assume that all the parameters are time independent.Time-dependent case can be solved in a similar way. To start with,we make the following transform
(t, x) = exp
2
22+
t+
2(x )
(t, x)
The modied Greens function solves the following propagationproblem
t (t, x) 12 22 (t, x) Z
0 (t, x +j) ejdj = 0
(0, x) = (x
) (t, 0) = 0, (t,) = 0
where =
2
, = 12 2 + +1
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 37 / 74
Greens function expansion
Perturbative calculation of Greens function in 1D. For brevity weassume that all the parameters are time independent
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
87/145
assume that all the parameters are time independent.Time-dependent case can be solved in a similar way. To start with,we make the following transform
(t, x) = exp
2
22+
t+
2(x )
(t, x)
The modied Greens function solves the following propagation
problem
t (t, x) 12 22 (t, x) Z
0 (t, x +j) ejdj = 0
(0, x) = (x
) (t, 0) = 0, (t,) = 0
where =
2
, = 12 2 + +1We assume that 1 and represent as follows
(t, x) = (0) (t, x) + (1) (t, x) + ...
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 37 / 74
Zero order term
A simple calculation yields (with = 2t)
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
88/145
A simple calculation yields (with = t)
(0) (t, x) =1p
n
x p
n
x + p
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 38 / 74
Zero order term
A simple calculation yields (with = 2t)
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
89/145
A simple calculation yields (with t)
(0) (t, x) =1p
n
x p
n
x + p
First order rhs
H(1) (t, x) = Px p
,p Px + p
,p
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 38 / 74
Zero order term
A simple calculation yields (with = 2t)
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
90/145
A simple calculation yields (with t)
(0) (t, x) =1p
n
x p
n
x + p
First order rhs
H(1) (t, x) = Px p
,p Px + p
,p
We use Duhamels principle and represent (1) as follows
(1) (t, x) =Zt
0
Z0(0) (t, x; s, 1) H(1) (s, 1) dsd1
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 38 / 74
First order term
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
91/145
A very elaborate calculation yields
(1) (t, x) =
21
P
x p
,p
+xPx + p ,p (x ) Px + p , p
(x + ) eP xp
,p
+ (x ) eP
xp
, p
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 39 / 74
First order term
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
92/145
A very elaborate calculation yields
(1) (t, x) =
21
P
x p
,p
+xPx + p ,p (x ) Px + p , p
(x + ) eP xp
,p
+ (x ) eP
xp
, p
The formula can be independently checked via the method of heatpotentials.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 39 / 74
Quality of approximation gure
pdf vs. x
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
93/145
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
0.000 1.000 2.000 3.000 4.000 5.000 6.000
x
pdf_ana pdf_exp_0 pdf_exp_1 pdf_num
Figure: G (t, x) for t = 20, = 20%, = 2%, = 4, = 1.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 40 / 74
Barrier PDF gure
sigma=0.2, lambda=0.1, alpha=1.0, mu=0.0, barrier=-2.0
sigma=0.2,lambda=0.1,
alpha=1 0
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
94/145
s g a 0 , a bda 0 , a p a 0, u 0 0, ba e 0
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
-5.00 -4.00 -3.00 -2.00 -1.00 0.00 1.00 2.00 3.00 4.00 5.00
X
Pro
b
Series PDE Laplace
alpha=1.0,
mu=0.0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 41 / 74
Hybrid model
Modeling of the value of the company is not always the best way ofd li i h d f l I i i h h h d li h k
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
95/145
dealing with default. It is sometimes thought that modeling the stock
price instead is a better choice.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 42 / 74
Hybrid model
Modeling of the value of the company is not always the best way ofd li ith d f lt It i ti th ht th t d li th t k
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
96/145
dealing with default. It is sometimes thought that modeling the stock
price instead is a better choice.
A typical model of this kind can be written as
dS = r (t) Sdt+ (t, S) dW (t)
where (t, 0) 6= 0. From that angle, one can argue that the Bacheliermodel is actually better than Black-Scholes!!
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 42 / 74
Hybrid model
Modeling of the value of the company is not always the best way ofd li ith d f lt It i ti th ht th t d li th t k
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
97/145
dealing with default. It is sometimes thought that modeling the stock
price instead is a better choice.A typical model of this kind can be written as
dS = r (t) Sdt+ (t, S) dW (t)
where (t, 0) 6= 0. From that angle, one can argue that the Bacheliermodel is actually better than Black-Scholes!!
Alternatively, one can add a terminal (Samuelson-style) jump todefault and model S as follows
dS = [r (t) + (S)] Sdt+ (t, S) dW (t) SdN(t)
Lipton and A Sepp (Bank of America Merrill L nch) Credit Modeling 02/03 42 / 74
Model Comparison
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
98/145
Comparison of numerical and analytical t.p.d.f. as well as the solutionof the barrier problem with non-constant barrier is given below.
A Li t d A S (B k f A i M ill L h) C dit M d li 02/03 43 / 74
Model Comparison
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
99/145
Comparison of numerical and analytical t.p.d.f. as well as the solutionof the barrier problem with non-constant barrier is given below.
By bootstrapping the barrier, we can reproduce term structure ofCDS for most names. In addition, we can price equity derivatives and
produce a respectable volatility skew.
A Li t d A S (B k f A i M ill L h) C dit M d li 02/03 43 / 74
Model Comparison
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
100/145
Comparison of numerical and analytical t.p.d.f. as well as the solutionof the barrier problem with non-constant barrier is given below.
By bootstrapping the barrier, we can reproduce term structure ofCDS for most names. In addition, we can price equity derivatives and
produce a respectable volatility skew.The barrier can assumed to be stochastic as in Due and Lando(2001) or CreditGrades (2000). Reduction of the information setmakes reduced form and structural modeling almost identical, Jarrow,Protter, Yildirim (2004).
A Li t d A S (B k f A i M ill L h) C dit M d li 02/03 43 / 74
Survival probability gure
100%
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
101/145
55%
70%
85%
0 1 2 3 4 5 6 7 8 9 10
RF MBC LFKY L AL
Figure: Typical survival probabilities Q(
0, T)
as functions of T (in years) for thereduced-form, Merton-Black-Cox, Lardy et al., Lipton, and Atlan-Lelanc models.Notice that QMBC (0, T) and QAL (0, T) are too at when T ! 0, whileQLFHY (0, 0) < 1. QRF (0, T) and QL (0, T) behave properly.
Li t d A S (B k f A i M ill L h) C dit M d li 02/03 44 / 74
Greens function without jumps
Without jumps, we need to nd Greens function for the correlated heatequation in the quarter-plane, i.e., to solve the following problem
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
102/145
t 12
21x1 x1 + 212x1x2 + 22x2 x2
+ 1x1 + 2x2 = 0
(t, x1, 0) = 0, (t, 0, x2) = 0
(t, x1, x2)
! (x1
1) (x2
2)
Standard transform
= exp (t+ 1 (x1 1) + 2 (x2 2))
removes drift terms provided that
= , = 121
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 45 / 74
Transforms
A sequence of linear transforms allows us to rewrite the pricing problem as
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
103/145
follows
(2)t
1
2
(2)
x(2)1 x
(2)1
+ (2)
x(2)2 x
(2)2
= 0
wt, x
(2)1 , 0
= 0, wt,x(2)2 /, x(2)2
= 0
w
t, x(2)
1 , x(2)2
!
x(
2)1 (2)1
x(2)
2 (2)2
where =p
12. We now have the standard heat equation in an angleand can use our previous results. This angle is formed by the horizontal
axis (2)
2 = 0 and a sloping line (2)
1 = (2)
2 /. It is acute when < 0and blunt otherwise. The size of this angle is =atan( /).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 46 / 74
Eigenfunction expansionThe solution of the above problem via the method of images wasindependently introduced in nance by He et al., Zhou and Lipton. Wewant to nd the Greens function for the following parabolic equation
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
104/145
want to nd the Green s function for the following parabolic equation
(2)t
1
2
(2)rr +
1
r
(2)r +
1
r2
(2)
supplied with the boundary conditions of the form
(2)
(t,
r,
) !r!0 C< , (2)
(t,
r,
) !r! 0, (2)
(t,
r,
0) = 0, (2)
and the initial condition
(2) (t, r, ) !t!0
(r r0) ( 0)r0
The fundamental solution has the form
t, r, j0, r0, 0 = 2e(r2 +r02)/2tt
n=1
In/
rr0
t
sin
n0
sin
n
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 47 / 74
Non-periodic Greens functionIt turns out that we can nd the fundamental solution via the method ofimages too. This approach seems to be new. To start with, we ndnon-periodic solution of the heat equation written in polar co-ordinates in
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
105/145
non periodic solution of the heat equation written in polar co ordinates in
the positive half-plane 0 < r < , < < :
t, r, j0, r0, 0 = 1 t, r, j0, r0, 02 t, r, j0, r0, 0where = 0, s =sign(),
1
t, r, j0, r0, 0 = e(r2 +r02)/2t2t
(s+ + s)2
e(rr0/t) cos
2 t, r, j0, r0, 0 = e
(r2 +r02)/2t
22
tZ
0
hs+e
(rr0/t) cosh((+)) + se(rr0/t) cosh(())
i2 + 1
d
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 48 / 74
Graph of Greens function
0.025
0.030
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
106/145
0.000
0.005
0.010
0.015
0.020
-20.00 -15.00 -10.00 -5.00 0.00 5.00 10.00 15.00 20.00
0.05 0.6 1.1 1.6 2.1 2.6 3.1 3.6 4.1 4.6
Figure: Cross-section of non-periodic Greens function. Parameters are as followst = 5, r0 = 1.5, 0 = 0.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 49 / 74
Method of images
Simple balancing of terms shows that
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
107/145
p g
2
t, r, j0, r0, 0 = O 02Next, we represent the fundamental solution in the form
t, r, j0, r0, 0 = n=
t, r, j0, r0, 0 + 2n t, r, j0, r0,0Indeed, it is clear that the sum converges, every term solves the parabolicequation, only one term has a pole inside the angle, and
(0) = () = 0.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 50 / 74
Correlated PDF gure
1.00
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
108/145
0.
00
0.
50
1.
00
1.
50
2.
00
2.
50
3.
00
3.
50
4.
00
4.
50
5.
00
5.
50
6.
00
6.
50
7.
00
7.
50
8.
00
8.
50
9.
00
9.
50
10.
00
0.00
1.20
2.40
3.60
4.80
6.00
7.20
8.40
9.60
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
Figure: PDF in the positive quarter-plane
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 51 / 74
Greens function with jumps
With jumps, we need to nd Greens function for the correlatedjump-diusion equation in the quarter-plane. In the simplest case we have
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
109/145
j p q q p p
to solve the following problem
t 12
21x1 x1 + 212x1 x2 + 22x2x2
+ 1x1 + 2x2
12Z
0
Z0 (x1 +j1, x2 +j2) e1
j12
j2 dj1dj2 + = 0
(t, x1, 0) = 0, (t, 0, x2) = 0
(t, x1, x2)
! (x1
1) (x2
2)
This problem has been recently solved (numerically) by Lipton and Sepp.Its analytical solution is not known (??).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 52 / 74
Multi-name case
Consider N companies and assume that their asset values are drivenby the following SDEs
dVi (t) J
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
110/145
dVi (t)
Vi (t) = (r di ii (t)) dt+ i (t) dWi (t) +
eJ
i 1 dNi (t)i = E
neJi 1
o
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 53 / 74
Multi-name case
Consider N companies and assume that their asset values are drivenby the following SDEs
dVi (t) J
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
111/145
dVi (t)
Vi (t) = (r di ii (t)) dt+ i (t) dWi (t) +
eJ
i 1 dNi (t)i = E
neJi 1
oDefault boundaries are driven by the ODEs of the form
dBi (t)Bi (t)
= (r di) dt Bi (0) = bi
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 53 / 74
Multi-name caseConsider N companies and assume that their asset values are drivenby the following SDEs
dVi (t) J
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
112/145
dVi (t)
Vi (t) = (r di ii (t)) dt+ i (t) dWi (t) +
eJ
i 1 dNi (t)i = E
neJi 1
oDefault boundaries are driven by the ODEs of the form
dBi (t)Bi (t)
= (r di) dt Bi (0) = biIn log coordinates
xi (t) = lnVi (t)
Bi (
t)
dxi (t) = i (t) dt+ i (t) dWi (t) + JidNi (t) xi (0) = ln
vi
bi
= i
i = 12 2i iiLipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 53 / 74
Multi-name caseConsider N companies and assume that their asset values are drivenby the following SDEs
dVi (t) J
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
113/145
dVi (t)
Vi (t) = (r di ii (t)) dt+ i (t) dWi (t) +
eJ
i 1 dNi (t)i = E
neJi 1
oDefault boundaries are driven by the ODEs of the form
dBi (t)Bi (t)
= (r di) dt Bi (0) = biIn log coordinates
xi (t) = lnVi (t)
Bi (
t)
dxi (t) = i (t) dt+ i (t) dWi (t) + JidNi (t) xi (0) = ln
vi
bi
= i
i = 12 2i iiLipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 53 / 74
Correlation structure
We introduce correlation between diusions in the usual way andassume that
dWi (t) dWj (t) = (t) dt
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
114/145
i ( ) j ( ) ij
( )
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 54 / 74
Correlation structure
We introduce correlation between diusions in the usual way andassume that
dWi (t) dWj (t) = ij (t) dt
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
115/145
i ( ) j ( ) ij ( )
We introduce correlation between jumps following the Marshall-Olkinidea. Let (N) be the set of all subsets of N names except for theempty subset f?g, and its typical member. With every weassociate a Poisson process N (t) with intensity (t), and
represent Ni (t) as follows
Ni (t) = 2(N)
1fi2gN (t)
i (t) = 2(N) 1fi2g (t)
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 54 / 74
Correlation structure
We introduce correlation between diusions in the usual way andassume that
dWi (t) dWj (t) = ij (t) dt
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
116/145
( ) j ( ) ij ( )
We introduce correlation between jumps following the Marshall-Olkinidea. Let (N) be the set of all subsets of N names except for theempty subset f?g, and its typical member. With every weassociate a Poisson process N (t) with intensity (t), and
represent Ni (t) as follows
Ni (t) = 2(N)
1fi2gN (t)
i (t) = 2(N) 1fi2g (t)
Thus, we assume that there are both common and idiosyncratic jumpsources.
A Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 54 / 74
Multi-name pricing problemWe now formulate a typical pricing equation in the positive cone
R(N)+ . We have
~(N)
~ ~
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
117/145
tU(t,
x) + L( )
U(t,
x) = (t,
x)U(t,~x0,k) = 0,k (t,~y) , U(t,~x,k) = ,k (t,~y)
U(T,~x) = (~x)
where ~x0,k, ~x,k, ~yk are N and N 1 dimensional vectors,respectively,
~x0,k =
x1, ..., 0
k, ...xN
, ~x,k =
x1, ...,
k, ...xN
, ~yk = (x1, ..., x
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 55 / 74
Multi-name pricing problemWe now formulate a typical pricing equation in the positive cone
R(N)+ . We have
~(N)
~ ~
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
118/145
tU(t,
x) + L U(t, x) = (t, x)U(t,~x0,k) = 0,k (t,~y) , U(t,~x,k) = ,k (t,~y)
U(T,~x) = (~x)
where ~x0,k, ~x,k, ~yk are N and N 1 dimensional vectors,respectively,
~x0,k =
x1, ..., 0
k, ...xN
, ~x,k =
x1, ...,
k, ...xN
, ~yk = (x1, ..., x
The integro-dierential operator L(N) can be written asL(N)f(~x) = 12
i
2i 2i f(~x) + i,j,j>i
aijijf(~x) +i
iif(~x)
f(~x) + 2(N)
i2
Jif(~x)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 55 / 74
Multi-name pricing problemWe now formulate a typical pricing equation in the positive cone
R(N)+ . We have
~(N)
~ ~
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
119/145
tU(t,
x) + L U(t, x) = (t, x)U(t,~x0,k) = 0,k (t,~y) , U(t,~x,k) = ,k (t,~y)
U(T,~x) = (~x)
where ~x0,k, ~x,k, ~yk are N and N 1 dimensional vectors,respectively,
~x0,k =
x1, ..., 0
k, ...xN
, ~x,k =
x1, ...,
k, ...xN
, ~yk = (x1, ..., x
The integro-dierential operator L(N) can be written asL(N)f(~x) = 12
i
2i 2i f(~x) + i,j,j>i
aijijf(~x) +i
iif(~x)
f(~x) + 2(N)
i2
Jif(~x)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 55 / 74
Jump terms
In the case of negative exponential jumps,
0
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
120/145
Jif(~x) =Zxi f(x
1, ..., xi +j, ...xN) i (j) dj
while in the case of discrete negative jumps
Jif(~x) = H (xi + Ji) f(x1, ..., xi + Ji, ...xN)
and H is the Heaviside function.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 56 / 74
Jump terms
In the case of negative exponential jumps,
0
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
121/145
Jif(~x) =Zxi f(x
1, ..., xi +j, ...xN) i (j) dj
while in the case of discrete negative jumps
Jif(~x) = H (xi + Ji) f(x1, ..., xi + Ji, ...xN)
and H is the Heaviside function.
We naturally split the operator L(N) into the local (dierential) andnon-local (integral) parts:
L(N)f(~x) = D(N)f(~x) +I(N)f(~x)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 56 / 74
Adjoint operator
The corresponding adjoint operator is
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
122/145
L(N)g(~x) = 12 i
2i 2i g (~x) +
i,j,j>i
aijijg (~x)i
iig(~x)
g (~x) + 2(N)
i2
Ji g(~x)
where
Ji g(~x) =Z
0g (x1, ..., xij, ...xN) i (j) dj
Ji g(~x) = g(x1, ..., xi
Ji, ...xN)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 57 / 74
Greens formulaWe introduce Greens function (t,~x), or, more explicitly,
t0,~x; t,~
, such that
~
( )
~
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
123/145
t0
t0,~xL(N) t0,~x = 0
t0,~x0k
= 0,
t0,~xk
= 0, (t,~x) = ~x~
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 58 / 74
Greens formulaWe introduce Greens function (t,~x), or, more explicitly,
t0,~x; t,~
, such that
~ L(N ) ~
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
124/145
t0
t0,~xL(N) t0,~x = 0
t0,~x0k
= 0,
t0,~xk
= 0, (t,~x) = ~x~
It can be shown that
U
t,~
=Z
R(N)+
(~x)
T,~x; t,~
d~x
+k
ZTt
ZR
(N1)+
k
t0,~y 1
22kk
t0,~y; t,~
dt0d~y
ZT
tZ
R(N)+
t0,~x
t0,~x; t,~
dt0d~xIn other words, instead of solving the backward pricing problem withnonhomogeneous rhs and boundary conditions, we can solve theforward propagation problem for Greens function with homogeneous
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 58 / 74
Single-name caseWe now are in a position to describe the relevant operators in moredetail for N = 1, 2.
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
125/145
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 59 / 74
Single-name caseWe now are in a position to describe the relevant operators in moredetail for N = 1, 2.When N = 1 we deal with the reference name alone. Since (1)
i f j l
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
126/145
consists of just one element,(1) = ff1gg
L(1)f = 12 2121f+ 11f f+ 1J1f = D(1)f+I(1)f
L(1)g = 1
22
12
1g
1
1
g
g + 1J
1g =
D(1)g +
I(1)g
= +1
I(1)f(x1) = 1Z0x1
f(x1 +j1) e1j1 d(1j1) = 1
Zx10
f(y1) e1 (x1y1 )
I(1)
g(x1) = 1Z0 g (x1 j1) e1
j1 d(1j1) = 1
Zx1
g(y1) e1 (x
1y
1 )
I(1)f(x1) = 1H (x1 + J1) f(x1 + J1)I(1)f(x1) = 1f(x1 J1)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 59 / 74
Two-name case 1When N = 2 we deal with the reference name 1 and protection seller2, say. Since (2) consists of three elements,
(2)
ff1g,f2g
,f1
,2gg
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
127/145
= ff1g f2g f1 2ggwe have
L(2)f = 12 2121f+ 12 2222f+ 1212 12f+ 11f+ 21f f+1
J1f+ 2
J2f+ 1,2
J1
J2f
= D(2)f+I(2)f
L(2)g = 12 2121g + 12 2222g + 1212 12g11g21g g+1
J
1 g + 2
J
2 g + 1,2
J
1
J
2 g
= D(2)g +I(2)g = +1 + 2 + 1,2
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 60 / 74
Two-name case 2
Specically,
I(2)f(x1, x2) = 1 Z
x1
0
f(y1, x2) e1 (x1y1 )d(1y1)
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
128/145
I Z0+2
Zx20
f(x1, y2) e2 (x2y2 )d(2y2)
+1,2
Zx10
Zx20
f(y1, y2) e1 (x1y1)2 (x2y2 )d(1y1
I(2)g(x1, x2) = 1Z
x1
g (y1, x2) e1 (x1y1 )d(1y1)
+2 Z
x2
g (x1, y2) e2 (x2y2 )d(2y2)
+1,2
Zx1
Zx2
g (y1, y2) e1 (x1y1 )2 (x2y2 )d(1y
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 61 / 74
Non-risky CDS pricing problemWe are now in a position to formulate the pricing problems of interest.
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
129/145
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 62 / 74
Non-risky CDS pricing problemWe are now in a position to formulate the pricing problems of interest.
We start with a CDS on a reference credit 1 which is sold by anonrisky seller to a nonrisky buyer. Assuming for simplicity that the
coupon is paid continuously at the rate s and that the default
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
130/145
coupon is paid continuously at the rate s, and that the defaultboundary is continuous, we write the pricing equation as follows
tU(1) (t, x1) +L(1)U(1) (t, x1) = (t, x1)
U(1)
(t, 0) = 0 (t) , U(1)
(t,) = (t)U(1) (T, x1) = 0
Here
(t, x1) = s
1 1 R1 e
1 x1 R1 =1R1
1 + 1
0 (t) = (1 R1) , (t) = sZT
tD
t, t0
dt0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 62 / 74
Risky seller CDS pricing problem 1
For a CDS on a reference credit 1 which is sold by a risky seller 2 to anonrisky buyer we write the pricing equation as follows
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
131/145
nonrisky buyer, we write the pricing equation as follows
tU(2) (t, x1, x2) + L(2)U(2) (t, x1, x2) = (t, x1, x2)
U(2) (t, 0, x2) = 0,
1
(t, x2) , U(2) (t,, x2) = ,
1
(t, x2)
U(2) (t, x1, 0) = 0,2 (t, x1) , U(2) (t, x1,) = ,2 (t, x1)
U(2) (T, x1, x2) = 0
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 63 / 74
Risky seller CDS pricing problem 2Here
(t, x1, x2) = s11 (t, x1, x2)22 (t, x1, x2)1,21,2 (t, x1, x2)1 (t x1 x2) =
1 R1
e1x1
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
132/145
1 (t, x1, x2) =
1 R1 e x2 (t, x1, x2) = U
(1) (t, x1) e2x2
1,2 (t, x1, x2) =
1 R1
e1 x1
1 e2 x2
+Z0x1
U(1) (t, x1 +j1) e1j1 d(1j1) e2x2 +
1 R1 R2e1x12 x2U(1) (t, x1) = R2U
(1)+ (t, x1) + U
(1) (t, x1)
0,
1 (t,
x2) = 1 R1,
,
1 (t,
x2) = sZT
t D
t,
t0
dt0
0,2 (t, x1) = R2U(1)+ (t, x1) + U
(1) (t, x1) , ,2 (t, x1) = U
(1) (t, x1)
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 64 / 74
Illustrations
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
133/145
s(0) L(0) R l(0) a(0) {disg {expg JPM 39.6 508 40% 203 243 0.1780 0.1780 5.6167 1.63
MS 29.31 534 40% 214 243 0.1286 0.1286 7.7759 2.41
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 65 / 74
Illustrations
CDS Spread Survival Prob Default Leg Annuity Leg
T JPM MS JPM MS JPM MS JPM
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
134/145
T JPM MS JPM MS JPM MS JPM1y 0.0055 0.0155 0.9909 0.9745 0.0055 0.0153 0.9954 0.98
2y 0.0055 0.0155 0.9818 0.9496 0.0109 0.0302 1.9818 1.94
3y 0.0055 0.0151 0.9729 0.9275 0.0163 0.0435 2.9591 2.88
4y 0.0055 0.0146 0.9640 0.9076 0.0216 0.0555 3.9276 3.80
5y 0.0059 0.0137 0.9518 0.8928 0.0289 0.0643 4.8858 4.70
6y 0.0060 0.0126 0.9417 0.8829 0.0350 0.0703 5.8323 5.59
7y 0.0060 0.0117 0.9323 0.8741 0.0406 0.0756 6.7693 6.47
8y 0.0060 0.0113 0.9230 0.8613 0.0462 0.0832 7.6969 7.33
9y 0.0060 0.0112 0.9138 0.8475 0.0517 0.0915 8.6153 8.1910y 0.0060 0.0110 0.9048 0.8339 0.0571 0.0997 9.5246 9.03
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 66 / 74
Illustrations
JPM0.04
$\lambda^{dis}$
$\lambda^{exp}$
Spread
MS
0.06
0.07
$\lambda^{dis}$
$\lambda^{exp}$
Spread
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
135/145
0.00
0.01
0.02
0.03
1y 2y 3y 4y 5y 6y 7y 8y 9y 10yT
0.00
0.01
0.02
0.03
0.04
0.05
1y 2y 3y 4y 5y 6y 7y 8y 9y 10yT
Figure: Calibrated intensity rates for JMP (lhs) and MS(rhs) in the models withENJs and DNJs, respectively.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 67 / 74
Illustrations
JPM
0.25
0.30JPM
0.25
0.30
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
136/145
0.00
0.05
0.10
0.15
0.20
0.00 0.10 0.20 0.30 0.40X
1y
5y
10y
0.00
0.05
0.10
0.15
0.20
0.00 0.10 0.20 0.30 0.40X
1y
5y
10y
Figure: PDF of the driver x(T) for JMP in the model with DNJs (lhs) and ENJs(rhs) for T = 1y, 5y, and 10 y.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 68 / 74
Illustrations
MS
0.16
0.18MS
0.16
0.18
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
137/145
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.00 0.10 0.20 0.30 0.40X
1y
5y
10y
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.00 0.10 0.20 0.30 0.40X
1y
5y
10y
Figure: Same graphs as in Figure 3 but for MS.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 69 / 74
Illustrations
JPM
50%
60% MS
50%
60%
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
138/145
0%
10%
20%
30%
40%
80% 100% 120% 140% 160% 180%K/S
CDSImpliedVol
$\lambda^{dis}$
$\lambda^{exp}$
0%
10%
20%
30%
40%
80% 100% 120% 140% 160% 180%K/S
CDSImpliedVol
$\lambda^{dis}$
$\lambda^{exp}$
Figure: Log-normal credit default swaption volatility implied from model withTe = 1y, Tt = 5y as a function of the inverse moneyness K/c(Te, Te + Tt).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 70 / 74
Illustrations
JPM
200%
250%
$\lambda^{dis}$
$\lambda^{exp}$
MS
200%
250%
$\lambda^{dis}$
$\lambda^{exp}$
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
139/145
0%
50%
100%
150%
14% 41% 69% 96% 123%K/S
EquityImpliedVol
MarketVol
0%
50%
100%
150%
30% 59% 89% 118%K/S
EquityImpliedVol
MarketVol
Figure: Log-normal equity volatility implied by the model as a function of inversemoneyness K/s for put options with T = 0.5y.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 71 / 74
Illustrations
rho=0.5
1.00
$\lambda^{dis}$
rho=0.99
1.00
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
140/145
0.00
0.20
0.40
0.60
0.80
0.25 1.25 2.25 3.25 4.25 5.25 6.25 7.25 8.25 9.25T
Gaussian
Correlation
$\lambda^{dis}$
$\lambda^{exp}$
0.00
0.20
0.40
0.60
0.80
0.25 1.25 2.25 3.25 4.25 5.25 6.25 7.25 8.25 9.25T
Gaussian
Correlation
$\lambda^{dis}$
$\lambda^{exp}$
Figure: Implied Gaussian correlations for FTDSs.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 72 / 74
Illustrations
146
156
InputSpread
Risky Seller, rho=0.75Risky Buyer, rho=0.75
Risky Seller, rho=0.055
60
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
141/145
96
106
116
126
136
1 2 3 4 5 6 7 8 9 10
T, years
Spread,
bp
y ,
Risky Buyer, rho=0.0
Risky Seller, rho=-0.75
Risky Buyer, rho=-0.75
30
35
40
45
50
1 2 3 4 5 6 7 8 9 10
T, years
Spread,
bp
InputSpread
Risky Seller, rho=0.75
Risky Buyer, rho=0.75
Risky Seller, rho=0.0
Risky Buyer, rho=0.0
Risky Seller, rho=-0.75
Risky Buyer, rho=-0.75
Figure: Equilibrium spread for protection buyer and protection seller of a CDS onMS with JPM as the counterparty (lhs); same for a CDS on JPM with MS as the
counterparty (rhs).
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 73 / 74
Conclusions
We gave a broad overview of the state of CDS modeling.
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
142/145
g g
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 74 / 74
Conclusions
We gave a broad overview of the state of CDS modeling.
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
143/145
We showed how to build meaningful structural default models andconnected them to reduced form models.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 74 / 74
Conclusions
We gave a broad overview of the state of CDS modeling.
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
144/145
We showed how to build meaningful structural default models andconnected them to reduced form models.
We showed that even the simplest elementary building blocks of
credit universe are dicult to describe properly.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 74 / 74
Conclusions
We gave a broad overview of the state of CDS modeling.W h d h b ild i f l l d f l d l d
http://find/http://goback/ -
8/3/2019 Lip Ton 2010
145/145
We showed how to build meaningful structural default models andconnected them to reduced form models.
We showed that even the simplest elementary building blocks of
credit universe are dicult to describe properly.The opinions expressed in this talk are those of the speaker and donot necessarily reect the views or opinions of Bank of AmericaMerrill Lynch.
Lipton and A Sepp (Bank of America Merrill Lynch) Credit Modeling 02/03 74 / 74
http://find/http://goback/