HAL Id: hal-00414280
https://hal.archives-ouvertes.fr/hal-00414280v2
Submitted on 29 Mar 2010
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Aurélien Alfonsi, Jérôme Lelong
To cite this version:
Aurélien Alfonsi, Jérôme Lelong. A closed-form extension to the Black-Cox model. International Journal of Theoretical and Applied Finance, World Scientific Publishing, 2012, 15 (8), pp.1250053:1- 30. �10.1142/S0219024912500537�. �hal-00414280v2�
Aurélien Alfonsi and Jérme Lelong
Université Paris-Est, CERMICS, Projet team MathFi ENPC-INRIA-UMLV, Eole des Ponts,
6-8avenue Blaise Pasal,77455MarneLa Vallée, Frane.
alfonsiermis.enp.fr
Eole Nationale Supérieure de Tehniques Avanées ParisTeh, Unité de Mathématiques Ap-
pliquées, 42bdVitor75015Paris, Frane.
LaboratoireJeanKuntzmann,UniversitédeGrenobleetCNRS,BP53,38041 GrenobleCédex9,
FRANCE.
e-mail : jerome.lelongimag.fr
Marh 29,2010
Abstrat
In theBlak-Coxmodel, a rm defaultswhen its value hits anexponential barrier. Here, we
propose an hybrid model that generalizes this framework. The default intensity an take two
dierent values and swithes whenthe rmvaluerosses a barrier. Ofourse, theintensity level
is higher belowthe barrier. We getan analyti formulafor the Laplae transform of thedefault
time. Thisresultan bealso extendedtomultiplebarriers andintensitylevels. Then, weexplain
howthis modelan be alibrated to Credit DefaultSwap pries andshow its tratability on dif-
ferent kindsof data. We also present numerial methods to numerially reover thedefault time
distribution.
Keywords: Credit Risk, Intensity Model, Strutural Model, Blak-Cox Model, Hybrid
Model, Parisian options, ParAsian options.
Aknowledgments. We would like to thank Jérme Brun and Julien Guyon from
Soiété Générale for providing us with market data and disussions. We also thank the
partiipantsof theonferene ReentAdvanementsinthe TheoryandPratieof Credit
Derivatives(Nie, September 2009) and espeially Monique Jeanblan, Alexander Lipton
and Claude Martini for fruitful remarks. Aurélien Alfonsi would like to aknowledge the
supportof the Chaire Risques Finaniers of Fondationdu Risque.
1 Introdution and model setup
Modelling rm defaults is one of the fundamental matter of interest in nane. It has
stimulated researh over the past deades. Clearly, the reent worldwide nanial risis
and its bunh of resounding bankrupties have underlined one again the need to better
understand redit risk. In this paper, we fous on the modelling of a single default. Usu-
ally, these models are divided into two main ategories: strutural and redued form (or
intensity) models.
Strutural models aim at explaining the default time with eonomi variables. In his
path breaking work, Merton [16℄ onneted the default of a rm with its ability to pay
bakits debt. The rm value isdened asthe sum ofthe equity value and thedebt value,
andis supposed tobeageometriBrownianmotion. Atthe bond maturity,defaultours
if the debtholders annot be reimbursed. In this framework, the equity value is seen as a
all option on the rm value. Then, Blak and Cox [5℄ have extended this framework by
triggeringthedefault assoonasthe rm value goesbelowsome ritialbarrier. Thus, the
default an our at any time and not onlyat the bond maturity. Many extensions of the
Blak Cox model, based on rst passage time, have been proposed in the literature. We
refer to the book of Bieleki and Rutkowski [4℄ for a nie overview. Reently, attention
has bepaid tothe alibrationof thesemodels toCreditDefault Swap(CDS inshort)data
(Brigo and Morini [7℄, Doreitner et al. [13℄). However, though eonomially sounded,
thesemodels anhardly beusedintensively onmarkets tomanageportfoliosespeiallyfor
hedging. Unless onsideringdynamiswith jumps(see Zhou[20℄forexample),theirmajor
drawbak is that the default time is preditable and no default an our when the rm
value islearly abovethe barrier. Inother words,they underestimatedefault probabilities
and redit spreads forshort maturities.
The prinipleof redued form modelsis todesribethe dynamis ofthe instantaneous
probability of default that is also alled intensity. This intensity is desribed by some
autonomous dynamis and the default event is thus not related to any riterion on the
solveny of the rm. We refer to the book of Bieleki and Rutkowski [4℄ for an overview
of these models. In general, they are designed for being easily alibrated to CDS market
data and are in pratie more tratable to manageportfolios.
However, none of these two kinds of model is fully satisfatory. In rst passage time
models, the default intensity is zero away from the barrier and the default event an be
foreast. Intensity models are in linewith CDS market data, but remain disonneted to
the rationales of the rm like its debt and equity values. Thus, they annot exploit the
informationavailable onequity markets. To overome this shortoming, and to provide a
unied framework for priing equity and redit produts, hybrid models have been intro-
dued, assuming that the default intensity is a (dereasing) funtion of the stok. Here,
we mention the works of Atlan and Leblan [2℄, and Carr and Linetsky [8℄ who onsider
the ase of a defaultableonstant elastiity model.
In this paper, we propose an hybrid model, whih extends the Blak-Cox model and
in whih the default intensity depends on the rm value. We present inthis introdution
framework when dealingwith redit risk and rm value models. Namely,we assume that
weareunderarisk-neutralprobabilitymeasurePandthatthe (riskless)shortinterestrate is onstant and equal to r > 0. We denote by (Ft, t ≥ 0) the default-free ltration and onsider a (Ft)-Brownian motion (Wt, t ≥ 0). We assume that the rm value (Vt, t ≥ 0)
evolvesaordingtotheBlak-Sholesmodelandthereforesatisesthefollowingdynamis:
dVt=rVtdt+σVtdWt, t ≥0, (1)
where σ > 0 is the volatility oeient. To model the default event, we assume that the
default intensity has the followingform:
λt =µ21{Vt<Ceαt} +µ11{Vt≥Ceαt}, (2)
where C > 0, α ∈ R, and µ2 > µ1 ≥ 0. This means that the rm has an instantaneous probability of default equalto µ2 or µ1 depending on whether its value is below or above
thetime-varyingbarrierCeαt. More preisely,letξdenoteanexponentialrandomvariable of parameter 1 independent of the ltration F. Then, we dene the default time of the
rm by:
τ = inf{t≥ 0, Z t
0
λsds≥ξ}. (3)
Asusual, wealsointrodue(Ht, t≥0)the ltrationgenerated bythe proess (τ∧t, t≥0)
and dene Gt = Ft ∨ Ht, so that (Gt, t ≥ 0) embeds both default-free and defaultable information.
This framework is a natural extension of the pioneering Blak-Cox model introdued
in [5℄, whih an indeed be simply seen as the limiting ase of our model when µ1 = 0
and µ2 →+∞. In the work of Blak and Cox, bankrupty an in additionhappen atthe
maturitydateofthebondsissuedbythermwhenthermvalueisbelowsomelevel. Here,
we donot onsider this possibility,even though itis tehnially feasible, beauseit would
make the default preditable in some ases. In the Blak-Cox model, the barrier Ceαt is
meanttobeasafetyovenantunderwhihdebtholdersanaskforbeingreimbursed. Here,
default an happen eitherabove orbelowthe barrier,whihrepresents insteadthe border
between two redit grades. Let us briey explain what typial parameter ongurations
ould be for this model. For a very safe rm, we expet that its value start above the
barrier with µ1 very lose to 0. The parameter µ2 should also be rather small sine it
annot be downgraded too drastially. Instead, for rms that are lose tobankrupty, we
expet tohave C < V0 and ahigh intensity ofdefault µ2. Then, the parametersshould be
suhthatthermisprogressivelydriftedtotheless riskyregion(i.e. r−σ2/2−α >0). In
fat, the CDSpries often reettwo possible outomes insuh ritialsituations. Either
the rm makesbankrupty inthe next future, oritsurvivesand isthen strengthened (see
Brigo and Morini [7℄for the Parmalat risis ase).
Now, we state the main theoretial result on whih this paper is based. It gives the
expliit formulafor the Laplae transformof the default time distribution.
Theorem 1.1. Let us set b = 1σlog(C/V0), m = σ1(r−α−σ2/2) and µb = µ21{b>0} + µ11{b≤0}. The default umulative distribution funtion P(τ ≤ t) is a funtion of t, b, m, µ1 and µ2 and is fully haraterized by its Laplae transform dened for z ∈ C+ := {z ∈ C,Re (z)>0},
Z ∞
0
e−ztP(τ ≤t)dt = emb−|b|√
2(z+µb)+m2
1
z+µ1 − 1 z+µ2
× (
−1{b>0} (4)
+ −m+p
2(z+µ2) +m2 p2(z+µ1) +m2+p
2(z+µ2) +m2 )
+1
z − 1
z+µb.
Theorem 1.1 an atually t in the framework of Theorem 2.1 with n = 2, where
the intensity an take n ≥ 2 dierent values instead of 2. Hene, we refer the reader to
Setion2for aproofof Theorem1.1, whihinfatomesfroma resultby Ka. Then,our
point ofview inthis paperistotakeadvantage of this resultand obtain afast alibration
proedureto CDS marketdata.
The Laplae transform (4) an also be obtained thanks to the results on Parisian
optionsbyChesneyetal.[11℄. This wasdoneinaformer version ofthispaperavailableon
http://hal.arhives-ouvert es.f r. The defaulttime τ, dened by(3), isrelatedtothe
time spent below and above the barrier. Other Blak-Cox extensions based on analytial
formulas for Parisian type options have been proposed in the reent past. Namely, Chen
and Suhaneki [10℄, Moraux [17℄ and Yu [19℄ onsidered the ase where the default is
triggered when the stok has spent a ertain amount of time in a row or not under the
barrier. Nonetheless, both extensions present the drawbak that the default is atually
preditable and the default intensity is either 0 or non-nite. This does not hold in our
framework.
The paper is strutured as follows. In Setion 2, we present the full modelfor whih
theintensityantaken ≥2dierentvaluesandweobtainalosedformulaforthe Laplae
transformofthe defaulttime. Setion3isdevotedtothepriingofCDSandstates simple
but interesting properties of the CDSspreads infuntion of the modelparameters. Then,
we fous on the alibration issue. Setion 4 is devoted to the alibration of the model
presented abovewhileinSetion5,wedisussthe alibrationofthe fullmodelwithn ≥3.
We present a general alibration proedure for the model and show ondierent pratial
settings how the model an t the market data. We nd our alibration results rather
enouraging. Last, we give in Setion 6 two methods to numerially invert the Laplae
transform of the default umulative distribution funtion given by Theorems 1.1 and 2.1.
Foreahmethod,westateinapreisewayitsauraywhihheavilyreliesontheregularity
of the funtion to be reovered. The required regularity assumptions are atually proved
tobesatised by the default umulative distributionfuntion in Appendix A.
2 The Laplae transform of the default distribution
Inthe introdution,wehaveonsideredadefaultintensitywhihtakestwodierentvalues
dependingonwhether the rm value isbelowor abovesome barrier. Here, wepresent the
full modelwhere the default intensity an take n≥2 dierent values,
0≤µ1 <· · ·< µn. (5)
We set C0 = +∞ and Cn = 0, and onsider C1, . . . , Cn−1 suh that Cn < Cn−1 < · · · <
C1 < C0. At time t ≥ 0, we assume that the default intensity of the rm is equal to µi,
when itsvalue isbetween Cieαt and Ci−1eαt. Thus, we set λt=
Xn
i=1
µi1{Cieαt≤Vt<Ci−1eαt}, (6)
and wedene the defaulttime τ exatlyasin (3). Assumption(5)means that thedefault
intensity is inreased (resp. dereased) eah time it rosses downward (resp. upward) a
barrier. Heuristially, these onstant intensities an be relatedto the redit grades of the
rm. Forarmindiulty,rossingdownwardthebarriersanalsorepresentthedierent
redit events that preede a bankrupty.
Now, we introdue notations that will be used throughout the paper. We set m = r−α−σ2/2 and
b0 = +∞, bi = 1
σlog(Ci/V0), i= 1, . . . , n−1 and bn =−∞. (7)
Thus, the default intensity (6) is equalto
λt= Xn
i=1
µi1{bi≤Wt+mt<bi−1}. (8)
From (3), we have
P(τ > t) =E h
e−
Rt
0
Pn
i=1µi1{
bi≤Ws+ms<bi−1}dsi
. (9)
Therefore,thedefaultdistribution(anditsLaplaetransform)onlydependonb= (bi)i=1,...,n−1,
m and µ= (µi)i=1,...,n. We set for t≥0 and z ∈C+
Pb,m,µ(t) = P(τ ≤t) and Pb,m,µc (t) =P(τ > t) = 1−Pb,m,µ, (10) Lb,m,µ(z) =
Z +∞ 0
e−ztP(τ ≤t)dt and Lcb,m,µ(z) = 1/z−Lb,m,µ(z), (11)
thatarerespetivelytheumulativedistributionfuntion,thesurvivalprobabilityfuntion
and their Laplaetransforms. When n= 2,we use the same notations as inthe introdu-
tion andsimplydenote by b= log(C1/V0)/σ the barrierlevel. We alsorespetively denote by Pb,m,µ1,µ2(t), Pb,m,µc 1,µ2(t), Lb,m,µ1,µ2(z) and Lcb,m,µ1,µ2(z) the quantities dened in (10)
and (11).
The following theorem gives a straightforward way to ompute the Laplae trans-
formLb,m,µ(z).
Theorem 2.1. In the above setting, Lb,m,µ(z) is given for z ∈C+ by
Lb,m,µ(z) = Xn
i=1
1{bi≤0<bi−1}
1
z − 1
z+µi −βi+−βi−
,
where R±(µ) = −m ± p
m2+ 2(z+µ). The oeients βi = [βi− βi+]′ are uniquely
determined by the indution:
βi = Πi−1β1+vi−1, i= 1, . . . , n (12)
and the onditions β1+ = βn− = 0. Here, Π0 = Id and Πi = Pi × · · · ×P1, v0 = 0 and vi =A−1(µi+1, bi)h
1
z+µi − z+µ1i+1 0i′
+Pivi−1 with:
Pi = 1
[R+(µi+1)−R−(µi+1)] × (13)
(R+(µi+1)−R−(µi)) ebi(R−(µi)−R−(µi+1)) (R+(µi+1)−R+(µi)) ebi(R+(µi)−R−(µi+1)) (R−(µi)−R−(µi+1)) ebi(R−(µi)−R+(µi+1)) (R+(µi)−R−(µi+1)) ebi(R+(µi)−R+(µi+1))
and
A−1(µi+1, bi) = 1
R+(µi+1)−R−(µi+1)
R+(µi+1) e−R−(µi+1)bi −e−R−(µi+1)bi
−R−(µi+1) e−R+(µi+1)bi e−R+(µi+1)bi
.
To solve the indution, one has rst to determine β1− by using that β1+ = βn− = 0
and (12) with i =n. Then, all the βi an be obtained with (12). When there is only one
barrier(i.e. n= 2), this an be solved expliitlyand the solution isgiven inTheorem 1.1.
Proof. We introdue for x∈R and t≥0, Xtx =x+Wt+mt, λ(x) =
Xn
i=1
µi1{xi≤x<xi−1} and p(t, x) = Eh
e−R0tλ(Xxs)dsi .
From (3)and (8), p(t,0) =P(τ > t) isthe survivalprobability funtion of τ.
Thanks to the Girsanov theorem, we have p(t, x) = e−mxE˜ h
emXtx−m2t/2e−R0tλ(Xsx)dsi
where Xtx is a Brownian motion starting from x under P˜. For z > 0, we onsider the
Laplae transformof p(t, x): x∈R, z >0, Lc(z, x) =
Z ∞
0
e−ztp(t, x)dt= e−mxE˜ Z ∞
0
e−(z+m2/2)temXtx−R0tλ(Xsx)dsdt
.
Now, from a result by Ka ([14℄, Theorem 4.9 p.271), it omes out that the Laplae
transform Lc(z, x) is C1 and pieewise C2 w.r.t. x,and solves:
∀i∈ {1, . . . , n}, bi ≤x < bi−1, 1−(z+µi)Lc(z, x) +m∂xLc(z, x) + 1
2∂x2Lc(z, x) = 0. (14)