Stochastic Di¤erential Equations Exercises
Exercise 11.1. The stochastic process
Ct=C0e Wt :t 0 ; r0 0
represents the exchange rate evolution, that isCtis the timetvalue in the domestic currency of one unit of the foreign currencyfWt:t 0g is a standard Brownian motion.
a) What is the stochastic di¤erential equation satis…ed byfCt:t 0g.
b) The price evolution of a risky asset, expressed in the foreign currency, is denoted fXt:t 0g. It satis…es the SDE
dXt= Xt dt+ Xt dWt
wherefWt :t 0gis a Brownian motion, independent offWt:t 0g:What is the SDE of the risky asset pricefYt :t 0g, expressed in the domestic currency?
Exercise 11.2.
a) Show that there is a unique solution to
dXt=a(b Xt)dt+ dW(t): b) Find explicitly this solution.
c) What behavior do you expect if the parameters are:.
a = 1; b= 1; = 1; t= 0;01and X0 = 1 a = 1; b= 1; = 1; t= 0;01 and X0 = 1 a = 1; b= 1; = 0;1; t= 0;01 and X0 = 1 a = 1; b= 2; = 0;3; t= 0;01 and X0 = 1
Exercise 11.3. Describe the behavior of the solution of dXt= 1
2 Xt dt+p
Xt(1 Xt)dWt
assuming that X0 is a random variable taken value in the unit interval [0;1]. Explain intuitively why Xt takes value between 0and 1.
Exercise 11.4. Assume that a particle move randomly in a plan such that its position at time t is given by Wt(1);Wt(2) where W(1) and W(2) are independent Brownian motions.
The distance to the origin is Bt =
r
Wt(1) 2+ Wt(2) 2; t >0:
Knowing that the quadratic covariation of two independent martingale is nil, use Itô’s lemma to show that
dBt= 1 2
1
Btdt+Wt(1)
Bt dWt(1)+ Wt(2)
Bt dWt(2):
Exercise 11.5. Show that the stochastic process fSt:t 0g where St=S0exp
Z t 0
sds
2
2 t+ Wt and W is a( ;F;fFt:t 0g; P) Brownian motion satis…es
dSt= tStdt+ StdWt:
Exercise 11.6. Vasicek model. Show that the stochastic process frt:t 0g where rt= +e t r0 + e t
Z t 0
e sdWs
and W is a( ;F;fFt:t 0g;P) Brownian motion satis…es drt = ( rt)dt+ dWt: Hint : Let Yt =Rt
0 e sdWs and show thatY is an Itô process.
Exercise 11.7. Consider the system of stochastic di¤erential equations:
dXt = ( Xt) dt+ dWt dYt = e e Yt dt+edfWt where the P Brownian motion W and fW are such that
CorrP h
Wt;fWti
= , 8t >0.
a) Find the unique strong solution of this system.
b) DetermineEP [Xt]:
c) The parameter is called the long term mean and the parameter is the speed of reversion. Justify these statements.
d) DetermineVarP[Xt]: e) Determine CorrP[Xt; Yt]:
Solutions
1 Exercise 11.1
a) Letf(w) =C0e w. Since d
dwf(w) =C0 e w and d2
dw2f(w) =C0 2e w; f(Wt) =Ct, df
dw(Wt) = Ct and d2f
dw2 (Wt) = 2Ct: Using Itô’s lemma, we get
dCt = Ct dWt+
2
2 Ct dt:
b) The multiplication rule leads to dYt = dXtCt
= Xt dCt+Ct dXt+dhX; Cit
= Xt Ct dWt+
2
2 Ct dt +Ct( Xt dt+ Xt dWt)
= Yt dWt+
2
2 Yt dt+ Yt dt+ Yt dWt
= Yt dWt+ Yt dWt +
2
2 + Yt dt:
2 Exercise 11.2
a) We need to verify that.
(i) jb(x; t) b(y; t)j+j (x; t) (y; t)j K jx yj; 8t 0 (ii) jb(x; t)j+j (x; t)j K (1 +jxj); 8t 0
(iii) E[X02]<1
Veri…cation of (i) :
jb(x; t) b(y; t)j+j (x; t) (y; t)j = ja(b x) a(b y)j+j j
= jaj jx yj:
Veri…cation of (ii) :
jb(x; t)j+j (x; t)j = ja(b x)j+ jabj+jaj jxj+
max (jabj+ ;jaj) (1 +jxj) LetK = max (jabj+ ;jaj).
Veri…cation of(iii) :because the initial condition is not speci…ed, we need to chose a square integrable random variable X0.
b) Let Zt=Xt b: Itô’s lemma implies that dZt = dXt
= a(b Xt)dt+ dW (t)
= Zt+ dW(t):
The stochastic process Z is an Ornstein-Uhlenbeck process and we have shown that the strong solution is
Zt=Z0e at+ e at Z t
0
easdWs: ReplacingZt with Xt b; we get
Xt b = (X0 b)e at+ e at Z t
0
easdWs:
That is why the solution Is
Xt=X0exp ( at) +b(1 exp ( at)) + exp ( at) Z t
0
exp (as)dWs: Let Yt =Rt
0 exp (as)dWs. Y =fYt:t2[0; T]g is an Itô process1 with Y0 = 0, Ks = 0 and Hs=eas. Moreover,
Xt =X0exp ( at) +b(1 exp ( at)) + exp ( at)Yt:
1Rappel. Le processus stochastiqueX =fXt: 0 t Tgest un processus d’Itô si pour toutt2[0; T]; Xt
P p.s.
= X0+ Z t
0
Ksds+ Z t
0
HsdWs
oùX0est une variable aléatoireF0 mesurable, les processusK=fKt: 0 t TgetH =fHt: 0 t Tg sontF adaptés,
Z T
0 jKsjds <1et Z T
0
Hs2ds <1P p.s.
Dans le cas qui nous occupe,RT
0 jKsjds=RT
0 0s= 0et RT
0 e2asds= 12e2T aa 1 <1:
Letf(y; t) = X0e at+b(1 exp ( at)) + e aty and note that
@f
@Y (y; t) = e at
@2f
@Y2 (y; t) = 0 d
dt X0e at+b(1 exp ( at)) + e aty = a X0e at+be at e aty
= a b X0e at b+be at e aty
= a(b f(y; t)):
Since Xt =f(Yt; t), the fourth version of Itô’s lemma2 gives the desired result:
df(Yt; t) = @f
@Y (Yt; t) Ht dWt + @f
@Y (Yt; t) Kt+@f
@t (Yt; t) + 1 2
@2f
@Y2 (Yt; t) Ht2 dt:
dXt = df(Xt; t)
= e at eat dWt+ e at 0 +a(b f(Yt; t)) + 1
20 (eas)2 dt
= dWt+a(b f(Yt; t)) dt
= dWt+a(b Xt) dt:
3 Exercise 11.3
The drift term 12 Xt induce a reversion behavior toward 12. Indeed, if Xt > 12, then
1
2 Xt < 0; implying that the local behavior of X tends to decrease just after t: On the other hand, if Xt< 12, then 12 Xt >0, implying that Xtends to increase. If Xt = 12, then the drift term is nil, having locally no impact on X.
The di¤usion termp
Xt(1 Xt)reaches a maximum atXt= 12. It worth zero whenever Xt= 0 or1. If Xt = 0, the drift term is positive and the di¤usion term is nil, implying that
2Rappel :
df(Yt; t) = @f
@Y (Yt; t) HtdWt + @f
@Y (Yt; t) Kt+@f
@t (Yt; t) +1 2
@2f
@Y2(Yt; t) Ht2 dt:
Xt+" is strictly positive. If Xt = 1, then the drift term is negative and the di¤usion term is nil, implying that Xt+" <1:
4 Exercise 11.4
Letf(w1; w2) =p
w21 +w22:
@f
@w1 (w1; w2) = w1
pw21+w22 = w1 f(w1; w2);
@f
@w2 (w1; w2) = w2
pw21+w22 = w2 f(w1; w2);
@2f
@w21 (w1; w2) = w22 pw12+w22
3 = w22
(f(w1; w2))3;
@2f
@w22 (w1; w2) = w21 pw12+w22 3
= w21 (f(w1; w2))3; and @2f
@w1@w2 (w1; w2) = w1w2 pw12+w22
3:
Therefore,
dBt = df Wt(1); Wt(2)
= Wt(1) f Wt(1); Wt(2)
dWt(1)+ Wt(2) f Wt(1); Wt(2)
dWt(2)
+1 2
Wt(2)
2
f Wt(1); Wt(2) 3
d W(1) t+1 2
Wt(1)
2
f Wt(1); Wt(2) 3
d W(2) t
Wt(1)Wt(2) f Wt(1); Wt(2)
3dD
W(1);Wt(2)E
t
= Wt(1) f Wt(1); Wt(2)
dWt(1)+ Wt(2) f Wt(1); Wt(2)
dWt(2)+ 1 2
Wt(1)
2
+ Wt(2)
2
f Wt(1); Wt(2) 3 dt
= Wt(1)
Bt dWt(1)+ Wt(2)
Bt dWt(2)+ 1 2
Bt2 Bt3dt
= Wt(1)
Bt dWt(1)+ Wt(2)
Bt dWt(2)+ 1 2
1 Btdt:
5 Exercise 11.7
a)
Xt = + (x0 ) exp ( t) + Z t
0
exp ( (t s)) dWs Yt = e+ x0 e exp ( et) +
Z t 0
eexp ( e(t s)) dfWs b)
EP [Xt] = EP + (x0 ) exp ( t) + Z t
0
exp ( (t s)) dWs
= + (X0 ) exp ( t) + EP Z t
0
exp ( (t s)) dWs
| {z }
=0
= + (x0 ) exp ( t):
c) If 0;
tlim!1EP[Xt] = lim
t!1 + (x0 ) exp ( t) = ; that is, 8" >0, 9t( ; ")>0such that 8t > t( ; "), EP[Xt] < ".
Note that if 0 1 < 2, then t( 1; ") > t( 2; "), that is, the larger is, more quickly EP[Xt] revert back to .
d)
VarP + (x0 ) exp ( t) + Z t
0
exp ( (t s)) dWs
= VarP Z t
0
exp ( (t s)) dWs
= EP
" Z t 0
exp ( (t s)) dWs
2#
= EP Z t
0
( exp ( (t s)))2 ds
= Z t
0
( exp ( (t s)))2 ds
=
2
2 1 e 2 t :
e)
CovP[Xt; Yt] = CovP
"
+ (x0 ) exp ( t) +Rt
0 exp ( (t s)) dWs; e+ x0 e exp ( et) +Rt
0 eexp ( e(t s)) dfWs
#
= CovP Z t
0
exp ( (t s)) dWs; Z t
0 eexp ( e(t s)) dfWs
= EP Z t
0
exp ( (t s)) dWs
Z t
0 eexp ( e(t s)) dfWs
= EP Z t
0
exp ( (t s))eexp ( e(t s)) ds
= e
Z t 0
exp ( ( +e) (t s)) ds
= e
+e 1 e ( +e)t :
CorrP[Xt; Yt] = CovP[Xt; Yt] q
VarP[Xt] q
VarP[Yt]
=
e
+e 1 e ( +e)t q 2
2 (1 e 2 t) qe2
2e(1 e 2et)
= 2 p e
+e
1 e ( +e)t
p(1 e 2 t) (1 e 2et):