Ornstein-Uhlenbek proesses
by
Marie Kaas Eriksen
THESIS
for the Masters degree of
Modelling and Data Analysis
(Master i modelleringogdataanalyse)
Faulty of Mathematis and Natural Sienes
University of Oslo
June 2008
Detmatematisk-naturvitenskapeligefakultet
Universiteteti Oslo
We study a mean-reverting model for interest rates. The model is an extension of the
Vasiekmodelandisasumofnon-GaussianOrnstein-Uhlenbekproesseswithsubordina-
tors,i.e. Lévyproesseswithonlypositivejumps,givingvariationoftheinterestrate. The
modelhavetheadvantagethatitgivesonlypositiveinterestrates,ontrarytotheVasiek
model. Wealulateexpliitresultsfortheharateristifuntionandtheautoorrelation
funtion oftheinterestrateforbothgeneralsubordinatorsand theasewherethesubor-
dinators are ompound Poisson. We alsond priesof zero-ouponbonds andEuropean
optionswritten on these bonds by applyingFouriermethods. It seems that themodel is
simpleenoughto allowfor analytialpriingof bondsand optionsin addition toapture
theharateristisoftheinterestrate. Intheend wedemonstratein asimulationhowthe
modelbehavewithertainvaluesofthevariables.
I would liketo thankmy patientsupervisor, FredEspen Benth, whoalwayshad timefor
metodisussanyproblem. Thethesisouldnothavebeenwrittenwithoutthehallenging
problemsandveryrelevantfeedbakhegaveme.
I would also like to thank family and friends, espeially my mother and father, who al-
wayssupported meduringmytimeasastudent.
Thankyou!
Oslo, May2008
MarieKaasEriksen
Abstrat i
Aknowledgements iii
1 Introdution 1
2 Some BasiTheory 3
2.1 MeasureTheoryandProbabilityTheory . . . 3
2.2 LévyProesses . . . 4
2.2.1 BrownianMotion . . . 5
2.2.2 LévyProesseswithJumps . . . 6
3 The Vasiek Model 9 3.1 SolutionandDistribution . . . 9
3.2 TheoretialAutoorrelationFuntion. . . 11
3.3 Zero-CouponBondPries . . . 12
4 Extension of the Vasiek Model 15 4.1 Solutionof
dX k (t)
andr(t)
. . . . . . . . . . . . . . . . . . . . . . . . . . . 154.2 CharateristiFuntion . . . 16
4.3 Moments . . . 21
4.4 TheoretialAutoorrelationFuntion. . . 25
4.5 ExpliitResultsforCompoundPoisson . . . 28
4.5.1 CharateristiFuntionof
r(t)
. . . . . . . . . . . . . . . . . . . . . 284.5.2 TheTheoretialAutoorrelationFuntionof
r(t)
. . . . . . . . . . . 355 Zero-Coupon BondPries 37 5.1 BondPries forGeneralSubordinators . . . 37
5.2 BondPries forCompoundPoissonSubordinators . . . 40
6 EuropeanBond Options 43 6.1 BondOptionPrieswith theVasiekModel . . . 43
6.2 BondOptionPrieswith ExtendedVasiekModel . . . 49
6.2.1 BondOptionPrieswithExtendedVasiekModel withGeneral
L k (t)
49 6.2.2 BondOptionPrieswithExtendedVasiekModelwhenL k (t)
isCom- pound Poisson . . . 537 Simulation 57
7.1 SimulationoftheInterestRate . . . 57
7.1.1 Simulationof
r(t)
intheVasiekModel . . . . . . . . . . . . . . . . 577.1.2 Simulationof
r(t)
intheExtendedVasiekModel . . . . . . . . . . 577.2 SimulationofZero-CouponBondPries . . . 60
7.2.1 SimulationofZero-CouponBondPries intheVasiek Model . . . 60
7.2.2 SimulationofZero-CouponBondPriesintheExtendedVasiekModel 60
A MATLAB Files 63
Bibliography 67
Introdution
Tomodelinterestratesandprieinterestratederivativesonbondsaregreatanddemand-
ing areasin mathematialnane. Interest ratederivativesare instrumentswhose payo
dependsontheleveloftheinterestrate. Thevolumeoftradingininterestratederivatives
inreasedinthe1980sand1990s. Thehallengeistondmodelsapturingtheharater-
istis of theinterest rate in a reasonabledegree. It's important that they're analytially
tratableaswell. TheVasiekmodelisoneoftherstmodelsofterm-strutureandisstill
anattrativelassofmodelsbeauseofitsanalytialproperties. Howeverithastheprop-
ertythattheinterestrateanbenegative. OthermodelsderivedaretheCox-Ingersoll-Ross
(CIR) model and theHull and White model, where the latterhas time-dependent oe-
ients. TheCIR model is anextension ofthe Vasiek model, but hasthe advantagethat
it only gives positive values. Some important interest rate derivatives are interest rate
aps/oors,swapoptionsand bond options. Wewill onlyinvestigatebondoptionsinthis
thesis.
In this thesis we disuss two models of the short-term interest rate. Therst oneis
theVasiekmodel,andtheother isanextensionoftheVasiekmodel,proposedin[2℄for
modelling spoteletriitypries. It'smotivatedfrom [1℄. Themodelis asumofOrnstein-
Uhlenbek(OU) proesses,eahwithapurejumpproesswithonlypositivejumps. Itts
well formodellingspot eletriitypries, beause theyareoften dependent oftheseason.
Sine the proess is a sum of OU proesses, it seemsreasonable to takeone of them to
model theseasonality. Inourase,withinterestrates,it hastheadvantagethatitseure
theinterestratetobepositive. Themodelisalsosimpleenough,suhthatoneanalulate
analytialexpressionsforommoninterestratederivatives.
Both of the models we onsider are mean-reverting. That a model is mean-reverting
meansthatitwilleventuallypullbakto someaveragelevel.
Themainpartofthethesisistoexaminethenewlassofmodelsdesribedabove. We
ndanautoorrelationfuntionoftheinterestrategivenbyasumofweightedexponentials
of a onstanttimes the time shift. We want to nd out how easy it is to obtain results
of zero-oupon bond pries and pries of European options written on these bonds. It
turns outthat wean easily deriveexpliit resultsfor the prieof zero-ouponbonds by
looking attheexpressionfortheinterestratediretly. TondpriesofEuropeanoptions
writtenonzero-ouponbondsaremoreompliatedthanndingpriesofEuropeanoptions
written onotherseurities. Thatis beauseinterestratesareused fordisountingaswell
asfor dening the payo from theoption. Wend the priebyapplying inverse Fourier
transform,andonean usefastFouriertransformtehniques toomputeitfurther.
The thesis is organized as follows: In hapter 2we give some well-known denitions
and resultsfrom measure and probabilitytheory. Wealso introdue stohasti proesses
likeBrownianmotionandpurejump proessesand statesomeoftheirproperties. All the
resultsaregivenwithoutproof. Inhapter3weonsidertheVasiekmodelandndpries
ofzero-ouponbondsandbondoptions. Weintroduethenewmodel,theextensionofthe
Vasiek model, in hapter 4. The rest of the thesisis dediated to the extended Vasiek
model. Wend the stationary harateristis, harateristi funtion and theorrelation
funtion. Inhapter5and6wederivepriesofzero-ouponbonds andEuropeanoptions
written on these bonds. Westate allour resultsbothin generaland forthe speial ase
whentheLévyproessesareompoundPoisson. Inhapter7wesimulatetheinterestrate
andpriesofzero-ouponbondswithmaturityinoneyear.
Appendix A ontainsthematlables usedto simulatetheinterestrateand thepries
ofzero-ouponbonds.
Some Basi Theory
Beforewestartlookingatourproblemweneedsomebasitheory. Thetheorystatedinthis
hapteriswell-knownandweskiptheproofs. Moreinformationandproofsanbefoundin
anybook instohastianalysis. Firstweintroduethenotionofa
σ
-algebra,aprobability measure and a probability spae. We state some well-known and useful theorems frommeasuretheory. Intheendwedene astohasti proessandlook atLévyproessesand
theirproperties.
2.1 Measure Theory and Probability Theory
Denition2.1 If
Ω
isagiven set,thenaσ
-algebraF
onΩ
isafamilyF
of subsetsofΩ
with
• ∅ ∈ F
.• F ∈ F ⇒ F c ∈ F
,whereF c = Ω \ F
.• F 1 , F 2 , · · · ∈ F ⇒ F =
∞
S
i=1
F i ∈ F
.Thepair
(Ω, F )
isalledameasurable spae.Denition2.2 Aprobability measureon
(Ω, F )
isafuntionP : F → [0, 1]
suhthat• P ( ∅ ) = 0
,P (Ω) = 1
.•
IfA 1 , A 2 , · · · ∈ F
and{ A i } ∞ i=1
isdisjoint, i.e.A i ∩ A j = ∅ , i 6 = j
,thenP
∞
[
i=1
A i
!
=
∞
X
i=1
P (A i ) .
Thetriple
(Ω, F , P )
isalled aprobability spae. Wealsodenethenotionof altration.Denition2.3 A ltration on a measurable spae
(Ω, F )
is an inreasing family ofσ
-algebras
{F t } ∈ F
suhthatF s ⊆ F t
,wheres ≤ t
.Altrationisinmathematialnaneusedtodesribetheinformationwegotuptilltoday.
Astimegoesby,weknowmoreandmore. Soithastobeinreasing.
Wemakethefollowingassumptionthroughoutthethesis:
Assumption2.1 All our models are modelled diretly under the risk-neutral probability
measure
Q
andthe probability spae we're working in, is(Ω, F , Q)
.Thenextthoremwillbeusedtoput thelimitoutsideexpetationswhen omputinghar-
ateristifuntions.
Theorem2.1 Bounded Convergene theorem
Let
µ(Ω) < ∞
. If thereexists a0 < k < ∞
suh that| f n | ≤ k µ
-a.e. andf n → f µ
-a.e.then
n→∞ lim Z
f n dµ = Z
f dµ,
and
n→∞ lim Z
| f n − f | dµ = 0.
Fubini's theoremallowsusto hangetheorderofintegrals.
Theorem2.2 Fubini's theorem
Let
(Ω i , F i , µ i )
,i = 1, 2
,beσ
-nitemeasurablespaes(i.e. thereexistaountableolletionofsets
A i 1 , A i 2 , . . . , ∈ F i
suhthat∪ n≥1 A i n = Ω
andµ i (A i n ) < ∞
foralln ≥ 1
andi = 1, 2
)andlet
f ∈ L 1 (Ω 1 × Ω 2 , F 1 × F 2 , µ 1 × µ 2 )
. ThenZ
Ω 1
Z
Ω 2
f dµ 2
dµ 1 =
Z
Ω 2
Z
Ω 1
f dµ 1
dµ 2 .
Wenowdenethenotionofastohastiproess.
Denition2.4 Astohastiproessisaparametrizedolletionofrandomvariables
{ X t }
,t ∈ T
denedona probability spae(Ω, F , P )
andassuming valuesinR n
.2.2 Lévy Proesses
TheLévyproessisanexampleofastohastiproess. It'snamedafterthemathematiian
PaulLévy. ALévyproess
L t
hasthefollowingproperties• L 0 = 0.
• L t
hasstationaryinrements,i.e. theprobabilitydistributionofanyinrementL t − L s
,dependsonlyonthelength
t − s
.• L t
hasindependentinrements,i.e. anytwonon-overlappinginrementsareindepen- dentofeah other.We willuse theharateristi funtion of aLévyproess, givenby theLévy-Khinhin
Theorem2.3 Lévy-Khinhin representation
Let
(X t )
beaLévyproessonR
withharateristitriplet(A, ν, γ)
. Then theharateristi funtion of(X t )
isE e izX t
= e ψ(z)t , z ∈ R d ,
where
ψ(z) = − 1
2 z.Az + iγ.z + Z
R
e izx − 1 − izx
1|x|≤1 ν(x)dx.
A
istheovarianematrixofaBrownianmotion,ν
istheLévymeasureandγ
isthedrift.2.2.1 Brownian Motion
A Brownian motion
B t
is an example of aLévy proess. It wasrst studied by RobertBrownin 1827. He wasstudying pollenpartiles oating in water under themirosope.
Brownianmotion is often usedbeause itmakesomputations simple, notbeause of its
auray. ItisaLévyproess,soitsatisesallthepropertiesabove,butitalsosatises
• B t − B s ∼ N (0, t − s)
.ForaBrownianmotion,theharateristitripletis
(1, 0, 0)
,so,fromLévy-Khinhinrepre- sentation, theharateristifuntion ofaBrownianmotionB(t)
beomesE h e izB(t) i
= e − 1 2 z 2 t
(2.1)A simple, but useful tool for solving stohasti dierential equations (SDE) is the It
formula.
Theorem2.4 The one - dimensional It formula.
Let
X t
bean Itproess wherethe dynamis isgiven bydX t = udt + vdB t .
Let
f (t, x) ∈ C 2 ([0, ∞ ] × R)
, i.e.f
is twotimesontinuously dierentiable on[0, ∞ ] × R
.Then
Y t = f (t, X t )
isagainan It proess,and
dY t = f t (t, X t )dt + f x (t, X t )dX t + 1
2 f xx (t, X t )(dX t ) 2 ,
where
f t (t, X t ) = ∂
∂t f (t, X t ), f x (t, X t ) = ∂
∂x f (t, X t ), f xx (t, X t ) = ∂ 2
∂x 2 F (t, X t ),
and
(dX t ) 2
isomputedaording tothe rulesdt · dt = dt · dB t = dB t · dt = 0, dB t · dB t = dt.
Proof. For proof,lookin Øksendal[4℄.
Anotherusefulproperty,whihwewillusetondthevarianeoftheVasiekmodel,is
Lemma2.1 It isometry.
E
Z T
S
f (t, ω)dB t
! 2
= E
"
Z T S
f 2 (t, ω)dt
# ,
for all
f
in the lassof funtionsg(t, ω) : [0, ∞ ) × Ω → R
suhthat
• (t, ω) → g(t, ω)
isB × F
-measurable,whereB
isthe Borelσ
-algebraon[0, ∞ )
.• g(t, ω)
isF t
-adapted,i.e.g(t, ω)
isF t
-measurablefor allt
.• E h R T
S g 2 (t, ω)dt i
< ∞
.In the setion disussion the priing of European bond options we make use of the
Girsanovtheoremto hangemeasure.
Theorem2.5 Girsanov's theorem
Let
B(t)
be astandardbrownian motion on aprobability spae(Ω, F , P)
. Supposethatγ u
isameasurableproess suhthat
E P h
e R 0 T γ u dB(u)− 1 2 R 0 T |γ u | 2 du i
= 1.
(2.2)Deneaprobability measure
P ˜
on(Ω, F T )
equivalenttoP
bymeansofthe Radon-Nikodým derivatived P ˜
dP = e R 0 T γ u dB(u)− 1 2 R 0 T |γ u | 2 du , P − a.s.
Then the proess
B(t) ˜
given bythe formulaB(t) = ˜ B(t) −
Z t 0
γ u du,
for all
t ∈ [0, T ]
,follows astandardBrownian motionon the spae(Ω, F , ˜ P)
.Asuientonditionfor(2.2)toholdis theNovikovondition:
E P h
e 1 2 R 0 T |γ u | 2 du i
< ∞ .
2.2.2 Lévy Proesses with Jumps
To ndan expliitrepresentationof
r(t)
in the extended Vasiekmodelwewill need theItformulaforsalarLévyproesses.
Proposition 2.1 It formula for salar Lévy proesses
Let
(X t ) t≥0
be a Lévy proess andf : R → R
aC 2
-funtion(i.e.2
timesdierentiable withontinuousderivatives). Thenf (X t ) = f (0) + Z t
0
σ 2 2
d 2
dx 2 f (X s )ds + Z t
0
d
dx f (X s− )dX s
+ X
0≤s≤t
∆X s 6=0
f (X s− + ∆X s ) − f (X s− ) − ∆X s
d
dx f (X s− )
.
Wearegoing tousethegeneralizedasewhere
f
alsodependsontime. Let(X t ) t≥0
beaLévyproessandlet
f : [0, T ] × R → R
beaC 1,2
-funtion(i.e1
timedierentiablein the rstvariableand2
timesdierentiablein theseond). Thenf (t, X t ) = f (0, X 0 ) + Z t
0
∂f
∂x (s, X s− )dX s + Z t
0
∂f
∂s (s, X s ) + σ 2 2
∂ 2 f
∂x 2 (s, X s )
ds
+ X
0≤s≤t
∆X s 6=0
f (s, X s− + ∆X s ) − f (s, X s− ) − ∆X s
∂f
∂x (s, X s− )
.
TheLévyproessesintheextendedVasiekmodelaresubordinators. Thatis,theyare
jump proesseswithonlypositivejumps. Theharateristitripletis then
(0, ν, 0)
. FromLévy-Khinhinrepresentationtheharateristifuntion ofsuhproessesis
E h e izL(t) i
= e ψ(z)t ,
(2.3)where
ψ(z) = R
R e izx − 1 − izx
1|x|≤1
ν (x)dx
.AnexampleofasubordinatoristheompoundPoissonproessandisdenedasfollows
Denition2.5 AompoundPoisson proesswith intensity
λ > 0
andjump sizedistribu-tion
f
isastohasti proessX t
denedasX t =
N t
X
i=1
Y i ,
where the jump sizes
Y i
areindependent and idential distributed(i.i.d.) with distributionf
,and(N t )
isaPoisson proess withintensityλ
,independent from(Y i ) i≥1
.The Vasiek Model
ThemodelanalysedbyVasiekin1977isoneoftherstmodelsoftermstruture. It has
somequalitiesthatmakesitattrative. Itislinearandanthereforebesolvedexpliitly. Its
distributionisGaussian,andzero-ouponbondsandotherderivativesareeasily obtained.
However,ahugedrawbakisthatitallowstheinterestratetobenegative.
TheVasiekmodeltakestheform
dr(t) = (µ − αr(t))dt + σdB(t).
(3.1)It's a mean-reverting Ornstein-Uhlenbek proess where
B(t)
is a Brownian motion andwhere
µ, α
andσ
arestritlypositiveonstants. Thattheproessismean-revertingmeans thatitwilleventuallypullbaktowardssomelong-runaveragelevel. Thatis,iftheinterest-rateis higherthan theexpeted, itwill tendtoderease, and ifitislower,itwill tendto
inrease.
3.1 Solution and Distribution
Thesolutionofthestohastidierentialequationaboveisgivenbythefollowingproposi-
tion.
Proposition 3.1 The solutionof (3.1)isgiven by
r(t) = r(s)e −α(t−s) + µ α
1 − e −α(t−s) + σ
Z t s
e −α(t−u) dB(u),
(3.2)wherethe proess startsattime
s ≤ t
.Proof. Toprovethepropositionwehaveto usetheItformula (Theorem2.4). If weuse
It'sformulaon
e αt r(t)
andinsertthedynamisofr(t)
from(3.1)wegetd(e αt r(t)) = αe αt r(t)dt + e αt dr(t) = e αt [µdt + σdB(t)].
So,
e αt r(t) − e αs r(s) = µ Z t
s
e αu du + σ Z t
s
e αu dB(u).
Andnallywegetthesolutionof
r(t)
:r(t) = r(s)e −α(t−s) + µ
α
1 − e −α(t−s) + σ
Z t s
e −α(t−u) dB(u)
Weannowndthedistributionof
r(t)
.Proposition 3.2 Theproess
r(t)
,givenby (3.2),isGaussiandistributedwithexpetationE [r(t)] = r(s)e −α(t−s) + µ α
1 − e −α(t−s)
andvariane
Var [r(t)] = σ 2 2α
1 − e −2α(t−s) .
Andwhentimegoestoinnityweget
t→∞ lim E [r(t)] = µ α
and
t→∞ lim Var [r(t)] = σ 2 2α .
Proof. Sine
e −α(t−u)
isdeterministi,wegetfromthepropertiesofBrownianmotionand It isometry thatσ R t
s e −α(t−u) dB(u)
is Gaussian with expeted value zero and varianegivenby
Var
σ Z t
s
e −α(t−u) dB(u)
= E
"
σ Z t
s
e −α(t−u) dB(u) 2 #
− E
σ Z t
s
e −α(t−u) dB(u) 2
= σ 2 Z t
s
e −2α(t−u) du = σ 2 2α
1 − e −2α(t−s) .
Itfollowsthattheproess
r(t)
isGaussiandistributedwithE [r(t)] = r(s)e −α(t−s) + µ α
1 − e −α(t−s)
and
Var [r(t)] = σ 2 2α
1 − e −2α(t−s) .
Findingthepropertiesof
r(t)
whentimegoestoinnityisstraightforward,t→∞ lim E [r(t)] = lim
t→∞
h r(s)e −α(t−s) + µ α
1 − e −α(t−s) i
= µ α
and
t→∞ lim Var [r(t)] = lim
t→∞
σ 2 2α
1 − e −2α(t−s)
= σ 2 2α .
Andtheproofisomplete.
3.2 The Theoretial Autoorrelation Funtion of
r ( t )
Wewanttoalulatethetheoretialautoorrelationfuntionfor
r(t)
. Theautoorrelation funtion says something about the degree of similarity betweenr(t)
and a time shiftedversionof itself. It an takevaluesin theinterval
[ − 1, 1]
,where1
means perfet positiveorrelation,and
− 1
meansperfet negativeorrelation.Theorrelationfuntionof r(t)isdenedby
corr (r(t), r(t + τ)) = E [(r(t) − E [r(t)]) (r(t + τ) − E [r(t + τ)])]
p Var [r(t)] Var [r(t + τ)]
= E [r(t)r(t + τ)] − E [r(t)] E [r(t + τ )]
p Var [r(t)] Var [r(t + τ)] .
Weomputethepartsseperately. Firstlookat
E [r(t)] E [r(t + τ )]
. Weusetheexpetationderivedin theprevioussetioninProposition (3.2).
E [r(t)] E [r(t + τ)]
=
r(s)e −α(t−s) + µ
α (1 − e −α(t−s) ) r(s)e −α(t+τ−s) + µ
α (1 − e −α(t+τ−s) )
= B.
Thenwelookat
E [r(t)r(t + τ)]
. Toalulatethispart,notiethatwefromthepropertiesofBrownianmotionhavethat
E h R t
s e −α(t−u) dB(u) i
iszero,sine
e −α(t−u)
isdeterministi.Rememberthat
r(t)
isgivenby(3.2).E [r(t)r(t + τ)] = E r(s)e −α(t−s) + µ
α (1 − e −α(t−s) ) + σ Z t
s
e −α(t−u) dB(u)
×
r(s)e −α(t+τ−s) + µ
α (1 − e −α(t+τ−s) ) + σ Z t+τ
s
e −α(t+τ−u) dB(u)
= B + σ 2 E Z t
s
e −α(t−u) dB(u) Z t+τ
s
e −α(t+τ−u) dB(u)
.
Theexpetationin thelast termoftheaboveequation anbeomputedas
E Z t
s
e −α(t−u) dB(u) Z t+τ
s
e −α(t+τ−u) dB(u)
= E Z t
s
e −α(t−u) dB(u) Z t
s
e −α(t+τ−u) dB(u) + Z t+τ
t
e −α(t+τ−u) dB(u)
= E
"
e −ατ Z t
s
e −α(t−u) dB(u) 2 #
+ E Z t
s
e −α(t−u) dB(u)
E Z t+τ
t
e −α(t+τ−u) dB(u)
= e −ατ E Z t
s
e −2α(t−u) du
= 1
2α e −ατ (1 − e −2α(t−s) ).
Hereweusedthe fat that
R t
s e −α(t−u) dB(u)
andR t+τ
t e −α(t+τ−u) dB(u)
are independentLet'sputitalltogether. Thenweget
E [r(t)r(t + τ)] − E [r(t)] E [r(t + τ)] = B + σ 2 2α e −ατ
1 − e −2α(t−s)
− B
= σ 2 2α e −ατ
1 − e −2α(t−s) .
Let'slookat
Var [r(t)] Var [r(t + τ)]
. WeusethevarianederivedintheprevioussetioninProposition3.2
Var [r(t)] Var [r(t + τ)] = σ 2 2α
1 − e −2α(t−s) σ 2 2α
1 − e −2α(t+τ−s)
= σ 4 4α 2
1 − e −2α(t−s) 1 − e −2α(t+τ−s) .
Proposition 3.3 The orrelation funtionof
r(t)
iscorr (r(t), r(t + τ)) = e −ατ 1 − e −2α(t−s) q
1 − e −2α(t−s)
1 − e −2α(t+τ−s) .
Whentime goes toinnitythe orrelationfuntion of
r(t)
tendstot→∞ lim corr (r(t), r(t + τ)) = e −ατ .
3.3 Zero-Coupon Bond Pries
Weareinterestedinndingthepriesofzero-ouponbonds,whereweassumethat
r(t)
ismodelled diretlyunder therisk-neutralprobability measure
Q
. A zero-ouponbondis abondpaying
1
urrenyat afuture timeT
withnoouponspaidinbetween.Denition3.1 Theprieof azero-ouponbondattime
t ≤ T
isP(t, T ) = E Q
h
e − R t T r(s)ds | F t i .
Tondtheprieweneedtoevaluate
− R T
t r(s)ds
. Asin (3.2),r(s)
isgivenbyr(s) = r(t)e −α(s−t) + µ
α
1 − e −α(s−t) + σ
Z s t
e −α(s−u) dB(u),
wheretheproessstartsattime
t ≤ s
.− Z T
t
r(s)ds = − Z T
t
r(t)e −α(s−t) ds − µ α
Z T t
1 − e −α(s−t) ds
− σ Z T
t
Z s t
e −α(s−u) dB(u)ds
= − I 1 − I 2 − I 3 .
Tomakeiteasier,weomputetheintegralsseperately. Westartwith
I 1
andI 2
. It'seasytoseethat
I 1 = r(t) Z T
t
e −α(s−t) ds = r(t) 1 α
1 − e −α(T −t)
(3.3)
I 2 = µ α
Z T t
1 − e −α(s−t) ds = µ
α (T − t) − µ α 2
1 − e −α(T −t)
.
(3.4)Thenlookat thelast integral
I 3
.I 3 = σ Z T
t
Z s t
e −α(s−u) dB(u) ds = σ Z T
t
e αu Z T
u
e −αs ds dB(u)
= σ Z T
t
1 α
1 − e −α(T −u) dB(u).
Hereweapplied Fubini'stheoremto hangetheorderoftheintegrals. Denenow
n(t, T ) = 1 α
1 − e −α(T−t) .
Thenweget
− Z T
t
r(s)ds = − r(t)n(t, T ) − µ 1
α (T − t) − 1 α n(t, T )
− σ Z T
t
n(u, T )dB(u).
Nowalulate
Z T t
n(u, T )du = Z T
t
1 α
1 − e −α(T −u) du = 1
α (T − t) − 1 α 2
1 − e −α(T −t)
= 1
α (T − t) − 1
α n(t, T ),
sowegetthat
− Z T
t
r(s)ds = − r(t)n(t, T ) − µ Z T
t
n(u, T )du − σ Z T
t
n(u, T )dB(u).
Tomakethenotationeasier,let
ξ T = − R T
t r(s)ds
.n(t, T )
isdeterministi,sowehavethatσ R T
t n(u, T )dB(u)
isGaussianwithexpetationzeroandvarianeσ 2 R T
t n 2 (u, T )du
byItisometry,Lemma2.1. Alsonotiethat
R T
t n(u, T )dB(u)
isindependentofF t
andthatr(t)
is
F t
-measurable. WethereforehaveP (t, T ) = E Q e ξ T |F t
= E Q e ξ T
= e −r(t)n(t,T )−µ R t T n(u,T)du E Q
h e −σ R t T n(u,T)dB(u) i
= e −r(t)n(t,T )−µ R t T n(u,T)du+ 1 2 σ 2 R t T n 2 (u,T)du .
Proposition 3.4 The zero-ouponbondprie, when
r(t)
isgiven by (3.2), isP(t, T ) = e m(t,T )−n(t,T)r(t) ,
where
n(t, T ) = 1 α
1 − e −α(T −t)
and
m(t, T ) = 1 2 σ 2
Z T t
n 2 (u, T )du − µ Z T
t
n(u, T )du.
Extension of the Vasiek Model
with Subordinators
Inthis hapter,and therest ofthethesis, wearegoing toinvestigatean extensionof the
Vasiekmodeloftheform
r(t) =
n
X
k=1
w k X k (t),
(4.1)where
dX k (t) = − α k X k (t)dt + dL k (t),
(4.2)and
X k (0) = r(0)
w 1
if
k = 1 0
ifk ≥ 2 .
Here,
L k (t), k = 1, 2, . . . , n
are subordinators. HavingseveralX k
's, rather thanjust one,gives an opportunityto apture dierent fators with inuene onthe interest rate
r(t)
.The model havethe advantagethat it always givespositiveinterest rates,something the
Vasiek model fails to do. It is, as mentioned before, proposed to model spot eletriity
priesin [2℄.
4.1 Solution of
dX k ( t )
andr ( t )
To nd an expliit solution of (4.2) we use the It formula for salar Lévy proesses,
Proposition2.1. ApplyingIt'sformulaon
f (t, X k (t)) = e α k t X k (t)
wegete α k t X k (t) − e α k s X k (s)
= Z t
s
e α k u dX k (u) + Z t
s
α k e α k u X k (u)du
+ X
s≤u≤t
∆X k (u)6=0
[e α k u (X k (u − ) + ∆X k (u)) − e α k u X k (u − ) − ∆X k (u)e α k u ]
= Z t
s
e α k u [ − α k X k (u)du + dL k (u)] + Z t
s
α k e α k u X k (u)du
= Z t
s
e α k u dL k (u).
Soanexpliitsolutionof (4.2)beomes
X k (t) = X k (s)e −α k (t−s) + Z t
s
e −α k (t−u) dL k (u),
(4.3)wheretheproessstartsatageneraltime
s ≤ t
. Wewantr(t)
tostarttoday,so sets = 0
.Ifweput
X k (t)
intotheexpressionforr(t)
wegetr(t) =
n
X
k=1
w k
X k (0)e −α k t + Z t
0
e −α k (t−u) dL k (u)
= w 1
r(0) w 1
e −α 1 t +
n
X
k=1
w k
Z t 0
e −α k (t−u) dL k (u)
= r(0)e −α 1 t +
n
X
k=1
w k
Z t 0
e −α k (t−u) dL k (u).
Proposition 4.1 An expliitsolution of
r(t)
startingattime0
,isr(t) = r(0)e −α 1 t +
n
X
k=1
w k
Z t 0
e −α k (t−u) dL k (u)
(4.4)4.2 The Charateristi Funtion of
r ( t )
We want to nd the harateristi funtion of
r(t)
. The harateristi funtion denes ompletelythe distributionof anyrandomvariable. Generally,theharateristifuntionofarandomvariable
X
isgivenbyϕ X (z) = E e izX
.
Sofor
r(t)
wehavetoomputeE e izr(t)
;
E h e izr(t) i
= e izr(0)e −α 1 t E h
e iz P n k=1 w k R 0 t e −αk (t−u) dL k (u) i .
First wetakealookat
E h
e iz P n k=1 w k R 0 t e −αk (t−u) dL k (u) i
. Todoso,let
f k (u) = e −α k (t−u)
.The
L k
'sareindependentofeahother, soE h
e iz P n k=1 w k R 0 t f k (u)dL k (u) i
=
n
Y
k=1
E h
e izw k R 0 t f k (u)dL k (u) i .
Let
{ u j } m j=1
be any partition of the interval [0, t
℄ withmax j | u j+1 − u j | < ǫ
. Then theintegralanbewrittenas
Z t 0
f k (u)dL k (u) = lim
ǫ→0 m
X
j=1
f k (u j )∆L k (u j ),
where
∆L k (u j ) = L k (u j+1 ) − L k (u j )
and,sinee g(t)
isaontinuousfuntion,wegetn
Y
k=1
E h
e izw k R 0 t f k (u)dL k (u) i
=
n
Y
k=1
E h
ǫ→0 lim e izw k P m j=1 f k (u j )∆L k (u j ) i .
The BoundedConvergene Theorem is applied to takethe limit outsidethe expetation.
Notiethat
∆L k (u j )
areindependentof∆L k (u j+1 )
forallj
sinetheLévyproesses,L k (u)
,haveindependentinrements. Thus
n
Y
k=1
E h
ǫ→0 lim e izw k P m j=1 f k (u j )∆L k (u j ) i
=
n
Y
k=1 ǫ→0 lim E h
e izw k P m j=1 f k (u j )∆L k (u j ) i
=
n
Y
k=1 ǫ→0 lim
m
Y
j=1
E h
e izw k f k (u j )∆L k (u j ) i .
Generally,wehavethatforaLévyproess
L(t)
,theharateristifuntionisE e izL(t)
= e ψ(z)t
, by Lévy-Khinhin representation (Theorem 2.3). It follows thatE
e iz∆L(t)
= e ψ(z)∆u
, whereψ(z) = R
R e izx − 1 − izx
1|x|≤1
ν(x)dx
, sine the proesses havehara-teristitriplet
(0, ν, 0)
. Finallyn
Y
k=1 ǫ→0 lim
m
Y
j=1
E h
e izw k f k (u j )∆L k (u j ) i
=
n
Y
k=1 ǫ→0 lim
m
Y
j=1
e ψ(zw k f k (u j ))∆u j
=
n
Y
k=1
ǫ→0 lim e P m j=1 ψ(zw k f k (u j ))∆u j
=
n
Y
k=1
e R 0 t ψ(zw k f k (u))du
= e P n k=1 R 0 t ψ(zw k f k (u))du .
Weputitalltogetherinaproposition.
Proposition 4.2 The harateristifuntion of
r(t)
isgiven byE h
e izr(t) i
= e izr(0)e −α 1 t e P n k=1 R 0 t ψ(zw k f k (u))du ,
(4.5)where
ψ(zw k f k (u)) = Z
R
e izw k f k (u)x − 1 − izw k f k (u)x
1|x|≤1
ν (x)dx
and
f k (u) = e −α k (t−u)
.Next we nd the expetation and variane of
r(t)
. We state the result in a proposition beforeweproveit.Proposition 4.3 The expetationandvarianeof
r(t)
aregiven byE [r(t)] = r(0)e −α 1 t +
n
X
k=1
w k
α k 1 − e −α k t Z
R
x − x
1|x|≤1
ν(x)dx,
(4.6)and
Var [r(t)] =
n
X
k=1
w k 2 2α k
1 − e −2α k t Z
R
x 2 ν(x)dx.
(4.7)Whentimegoestoinnitytheybeome
t→∞ lim E [r(t)] =
n
X
k=1
w k α k
Z
R
x − x
1|x|≤1 ν (x)dx
and
t→∞ lim Var [r(t)] =
n
X
k=1
w k 2 2α k
Z
R
x 2 ν(x)dx.
Intheproof wewill makeuseof thefollowingorollary. We stateit withoutproofbefore
provingProposition4.3.
Corollary4.1 If
X
isarandomvariablewithharateristifuntionϕ X (z) = E[e izX ]
oneannd it's
n
'thmoment byusing the formulaE[X n ] = (i) −n d n
dz n E [ϕ X (z)]
z=0 .
Proof of Proposition4.3. Westartwiththeexpetation. Notiethat
L k
isindependentofL j
whenk 6 = j
. Rememberthatr(t)
isgivenby(4.4).E [r(t)] = E
"
r(0)e −α 1 t +
n
X
k=1
w k Z t
0
e −α k (t−u) dL k (u)
#
= r(0)e −α 1 t +
n
X
k=1
w k E Z t
0
e −α k (t−u) dL k (u)
.
Wehavetond
E h R t
0 e −α k (t−u) dL k (u) i
. Todothenotationeasier,let
f k (u) = e −α k (t−u)
.From the alulations leading up to Proposition 4.2, we know that the harateristi
funtion of
R t
0 f k (u)dL k (u)
, withψ(zf k (u)) = R
R e izf k (u)x − 1 − izf k (u)x
1|x|≤1
ν (x)dx
,is
e R 0 t ψ(zf k (u))du
. SobyCorollary4.1E
Z t 0
f k (u)dL k (u)
= − i d dz E h
e iz R 0 t f k (u)dL k (u) i z=0
= − i d
dz e R 0 t ψ(zf k (u))du z=0
= − i d
dz Z t
0
ψ(zf k (u))du
e R 0 t ψ(zf k (u))du z=0
= − i Z t
0
d
dz ψ(zf k (u))du e R 0 t ψ(zf k (u))du z=0
,
with
ψ(zf k (u))
asabove. Wehavethatψ(0) = 0
,soe R 0 t ψ(zf k (u))du | z=0 = 1
. FurtherE
Z t 0
f k (u)dL k (u)
= − i Z t
0
d
dz ψ(zf k (u))du z=0
(4.8)
Wethenneedtoompute
d
dz ψ(zf k (u)) | z=0
.d
dz ψ(zf k (u)) z=0
= Z
R
if k (u)xe izf k (u)x − ixf k (u)
1|x|≤1
ν(x)dx z=0
= Z
R
if k (u)x − ixf k (u)
1|x|≤1
ν(x)dx.
Weget
E Z t
0
f k (u)dL k (u)
= 1
α k 1 − e −α k t Z
R
x − x
1|x|≤1
ν (x)dx,
(4.9)whih isaresultofthefollowingomputation
E Z t
0
f k (u)dL k (u)
= − i Z t
0
Z
R
if k (u)x − ixf k (u)
1|x|≤1
ν (x)dx
du
= Z t
0
f k (u) Z
R
x − x
1|x|≤1
ν(x)dx
du
= Z
R
x − x
1|x|≤1 ν (x)dx
Z t 0
f k (u)du
= Z
R
x − x
1|x|≤1 ν (x)dx
Z t 0
e −α k (t−u) du
= Z
R
x − x
1|x|≤1
ν (x)dx 1 α k
1 − e −α k t .
Ifweputitalltogether,wegetthat
E [r(t)] = r(0)e −α 1 t +
n
X
k=1
w k α k
1 − e −α k t Z
R
x − x
1|x|≤1
ν(x)dx.
Thisproves(4.6). Lettingtimegoto innity,
lim t→∞ e −α k t = 0
givest→∞ lim E [r(t)] =
n
X
k=1
w k
α k
Z
R
x − x
1|x|≤1
ν (x)dx.
Itremainstoshowthatthevarianeisgivenby(4.7). Let
f k (u)
stillbegivenbye −α k (t−u)
.Notiethat
L k
isindependentofL j
whenk 6 = j
. WethenndVar [r(t)] = Var
"
r(0)e −α 1 t +
n
X
k=1
w k
Z t 0
f k (u)dL k (u)
#
= Var
" n X
k=1
w k
Z t 0
f k (u)dL k (u)
#
=
n
X
k=1
w 2 k Var Z t
0
f k (u)dL k (u)
.
We want to ompute
Var h R t
0 f k (u)dL k (u) i
. Generally, we have that
Var[X ] = E[X 2 ] − E[X ] 2
. SoVar h
R t
0 f k (u)dL k (u) i
= E
R t
0 f k (u)dL k (u) 2
− E h R t
0 f k (u)dL k (u) i 2
. Compute
the
n
'thmomentofarandomvariableX
byusingtheformulaE[X n ] = (i) −n d dz n n ϕ X (z) | z=0
,where
ϕ X (z)
istheharateristifuntionofX
. Wehavealreadyalulatedtheharater-istifuntionof
R t
0 f k (u)dL k (u)
,whendealingwiththeharateristifuntionofr(t)
beforeProposition4.2. Hene
E
" Z t 0
f k (u)dL k (u) 2 #
= (i) −2 d 2 dz 2 E h
e iz R 0 t f k (u)dL k (u) i z=0
= − d 2
dz 2 e R 0 t ψ(zf k (u))du z=0
= − d dz
Z t 0
d
dz ψ(zf k (u))du e R 0 t ψ(zf k (u))du z=0
= −
"
Z t 0
d
dz ψ(zf k (u))du 2
e R 0 t ψ(zf k (u))du
+ Z t
0
d 2
dz 2 ψ(zf k (u))du e R 0 t ψ(zf k (u))du
# z=0
= −
" Z t 0
d
dz ψ(zf k (u))du 2
+ Z t
0
d 2
dz 2 ψ(zf k (u))du
# z=0
,
where weused that
e 0 = 1
. We havethatR t 0
d
dz ψ(zf k (u))du
z=0 = i E h R t
0 f k (u)dL k (u) i
from(4.8). Rememberthat
E h R t
0 f k (u)dL k (u) i
isgivenby(4.9). Wethenget
Z t 0
d
dz ψ(zf k (u))du z=0
2
= − 1
α 2 k 1 − e −α k t 2 Z
R
x − x
1|x|≤1
ν (x)dx 2
Soit remains to ompute
R t 0
d 2
dz 2 ψ(zf k (u))du | z=0
. Werst omputethe expression insidetheintegral;
d 2
dz 2 ψ(zf k (u)) z=0
= d 2 dz 2
Z
R
e izf k (u)x − 1 − ixzf k (u)
1|x|≤1
ν(x)dx
z=0
= Z
R
d dz
h
if k (u)xe izf k (u)x − ixf k (u)
1|x|≤1
ν (x) i
dx z=0
= − Z
R
f k 2 (u)x 2 ν (x)dx.
Rememberthat
f k (u) = e −α k (t−u)
. It followsthatZ t
0
d 2
dz 2 ψ(zf k (u))du z=0
= − Z t
0
Z
R
f k 2 (u)x 2 ν(x)dx
du
= − Z
R
x 2 ν(x)dx Z t
0
f k 2 (u)du
= − Z
R
x 2 ν(x)dx Z t
0
e −2α k (t−u) du
= − Z
R
x 2 ν(x)dx 1 2α k
1 − e −2α k t
Goingbaktoourexpressionfor
E
R t
0 f k (u)dL k (u) 2
,andusewhatwehavefound,we
get
E
"
Z t 0
f k (u)dL k (u) 2 #
= 1
α 2 k 1 − e −α k t 2 Z
R
x − x
1|x|≤1
ν (x)dx 2
+ 1 2α k
1 − e −2α k t Z
R
x 2 ν(x)dx.
From(4.9)itfollowsdiretlythat
E Z t
0
f k (u)dL k (u) 2
= 1
α 2 k 1 − e −α k t 2 Z
R
x − x
1|x|≤1 ν (x)dx
2
.
Tomaketheproofof (4.7)omplete;
Var [r(t)] =
n
X
k=1
w k 2 Var Z t
0
f k (u)dL k (u)
=
n
X
k=1
w k 2 E
" Z t 0
f k (u)dL k (u) 2 #
− E Z t
0
f k (u)dL k (u) 2 !
=
n
X
k=1
w k 2
"
1
α 2 k 1 − e −α k t 2 Z
R
x − x
1|x|≤1 ν(x)dx
2
+ 1 2α k
1 − e −2α k t Z
R
x 2 ν(x)dx
#
−
n
X
k=1
w k 2
α 2 k 1 − e −α k t 2 Z
R
x − x
1|x|≤1 ν(x)dx
2
=
n
X
k=1
w k 2 2α k
1 − e −2α k t Z
R
x 2 ν (x)dx.
Wealsohavetohekwhathappenswhentimegoestoinnity;
t→∞ lim Var [r(t)] =
n
X
k=1
w 2 k 2α k
Z
R
x 2 ν(x)dx.
Andourproofis omplete.
4.3 Moments of
r ( t )
We wantto nd the rsttwo momentsof
r(t)
. As mentionedbefore, wean dothat byusingCorollary4.1. Therstmomentisalreadyomputedandgivenby(4.6). Westateit
Proposition 4.4 The rstmoment of
r(t)
isE [r(t)] = r(0)e −α 1 t +
n
X
k=1
w k
α k
1 − e −α k t Z
R
x − x
1|x|≤1
ν(x)dx.
Tondtheseondmomentwelookat
E r 2 (t)
= (i) −2 d 2 dz 2 E h
e izr(t) i z=0
= (i) −2 d dz
d dz E h
e izr(t) i z=0
.
Let
f k (u) = e −α k (t−u)
asbefore. First,letusomputedz d 2 2 E e izr(t)
. Weusetheexpression
for theharateristifuntion of
r(t)
given by Proposition 4.2. Letb(t) = r(0)e −α 1 t
andrememberthat
ψ(zw k f k (u)) = Z
R
e izw k f k (u)x − 1 − izw k f k (u)x
1|x|≤1
ν(x)dx.
(4.10)Thenweget
d dz E h
e izr(t) i
= d dz
h
e izb(t) e P n k=1 R 0 t ψ(zw k f k (u))du i
= ib(t)e izb(t)
n
Y
k=1
e R 0 t ψ(zw k f k (u))du + e izb(t) d dz
n
Y
k=1
e R 0 t ψ(zw k f k (u))du ,
and
d 2 dz 2 E h
e izr(t) i
= − b 2 (t)e izb(t)
n
Y
k=1
e R 0 t ψ(zw k f k (u))du + ib(t)e izb(t) d dz
n
Y
k=1
e R 0 t ψ(zw k f k (u))du
+ib(t)e izb(t) d dz
n
Y
k=1
e R 0 t ψ(zw k f k (u))du + e izb(t) d 2 dz 2
n
Y
k=1
e R 0 t ψ(zw k f k (u))du .
Now,let
D 1 = − b 2 (t)e izb(t)
n
Y
k=1
e R 0 t ψ(zw k f k (u))du
z=0
(4.11)D 2 = ib(t)e izb(t) d dz
n
Y
k=1
e R 0 t ψ(zw k f k (u))du
z=0
(4.12)D 3 = e izb(t) d 2 dz 2
n
Y
k=1
e R 0 t ψ(zw k f k (u))du
z=0 .
(4.13)Notiethat
E r 2 (t)
= (i) −2 [D 1 + 2D 2 + D 3 ] .
(4.14)Weomputethepartsseparately,but notierstthat
e R 0 t ψ(zw k f k (u))du
z=0 = 1,
(4.15)sine