Stochastic Differential Equationsocw.nctu.edu.tw/upload/classbfs1209013157139887.pdf · CHAPTER 10...
Click here to load reader
Transcript of Stochastic Differential Equationsocw.nctu.edu.tw/upload/classbfs1209013157139887.pdf · CHAPTER 10...
CHAPTER 10
Stochastic Differential Equations
Consider a stochastic process (Xt) satisfying
dXt = b(t,Xt,Wt) dt + σt, Xt,Wt) dWt. (10.1)
Question.
(1) Can we obtain the existence and uniqueness theorem for (10.1)? What are the
properties of the solution?
(2) How to solve it?
10.1. Examples and some solution methods
Example 10.1. Solving
dXt = αXt dWt + σXt dt, (10.2)
where the initial condition X0 is given and α, σ are constant. Rewrite this equation as
dXt
Xt
= α dWt + σ dt,
and we can get ∫ t
0
dXu
Xu
=
∫ t
0
α dWu +
∫ t
0
σ du = αWt + σt. (10.3)
Due to the Ito’s formula,
df(t,Xt) =∂f
∂tf(t,Xt) dt +
∂f
∂xf(t,Xt) dXt +
1
2
∂2f
∂x2f(t,Xt) (dXt)
2,
237
238 10. STOCHASTIC DIFFERENTIAL EQUATIONS
we may take f(t, x) = f(x) = ln x, then
f ′(x) =1
x, f ′′(x) = − 1
x2.
Then
d ln(Xt) =1
Xt
dXt − 1
2
1
X2t
(dXt)2 =
1
Xt
dXt − α2
2dt,
i.e.,
ln(Xt) − ln(X0) =
∫ t
0
dXu
Xu
− α2t
2. (10.4)
Combining (10.3) and (10.4), we get
ln
(Xt
X0
)+
α2t
2=
∫ t
0
dXu
Xu
= αWt + σt.
Thus, the solution to (10.2) is given by
Xt = X0 exp
(αWt +
(σ − α2
2
)t
).
Definition 10.2. A stochastic process (Xt) of the form
Xt = X0 exp(αWt + μt)
is called the geometric Brownian motion.
Remark 10.3. (1) If (Wt) is independent of X0, then
E[Xt] = E[X0 exp
(αWt +
(σ − α2
2
)t
)]= E[X0]e
σt.
(2) (i) If σ > α2/2, then Xt −→ ∞ as t → ∞ P-a.s.
(ii) If σ < α2/2, then Xt −→ 0 as t → ∞ P-a.s.
(iii) If σ = α2/2, then Xt will fluctuate between arbitrary large and arbitrary
small values as t → ∞ P-a.s.
10.1. EXAMPLES AND SOME SOLUTION METHODS 239
Example 10.4 (Hull-White interest-rate model). Consider
dRt = (at − btRt) dt + σt dWt, with R0 = r, (10.5)
where at, bt, and σt are deterministic function. Then
dRt + btRt dt = at dt + σt dWt,
which implies that
d
(Rt exp
(∫ t
0
bu du
))= at exp
(∫ t
0
bu du
)dt + σt exp
(∫ t
0
bu du
)dWt.
Thus, the solution to (10.5) is given by
Rt = r exp
(−
∫ t
0
bu du
)+
∫ t
0
as exp
(−
∫ t
s
bu du
)ds
+
∫ t
0
σs exp
(−
∫ t
s
bu du
)dWs.
Example 10.5. Consider the stochastic differential equation
dXt = rXt(K − Xt) dt + βXt dWt, with X0 = x > 0. (10.6)
Rewrite the equation as
dXt
Xt
+ rXt dt = rK dt + β dWt.
Taking integration on the both sides, we have
∫ t
0
dXu
Xu
+ r
∫ t
0
Xu du = rKt + βWt.
Using a similar argument as in Example 10.1, we get
∫ t
0
dXu
Xu
= ln(Xt) − ln(X0) +1
2
∫ t
0
1
X2u
d〈X〉u
= ln
(Xt
x
)+
1
2
∫ t
0
1
X2u
β2X2u du = ln
(Xt
x
)+
1
2β2t.
240 10. STOCHASTIC DIFFERENTIAL EQUATIONS
Thus,
ln
(Xt
x
)+ r
∫ t
0
Xu du = βWt +
(rK − 1
2β2
)t,
which implies
Xt exp
(r
∫ t
0
Xu du
)= x exp
[βWt +
(rK − 1
2β2
)t
].
Integration with respect to t on the both sides, we obtain
x
∫ t
0
exp
[βWs +
(rK − 1
2β2
)s
]ds =
∫ t
0
Xs exp
(r
∫ s
0
Xu du
)ds
=
∫ t
0
exp
(r
∫ s
0
Xu du
)d
(∫ s
0
Xu du
)
=1
rexp
(r
∫ s
0
Xu du
)∣∣∣∣t
s=0
=1
r
[exp
(r
∫ t
0
Xu du
)− 1
].
Hence,
exp
(r
∫ t
0
Xu du
)= 1 + rx
∫ t
0
exp
[βWs +
(rK − 1
2β2
)s
]ds,
i.e., ∫ t
0
Xu du =1
rln
(1 + rx
∫ t
0
exp
[βWs +
(rK − 1
2β2
)s
]ds
).
Taking derivative with respect to t, we see that the solution to (10.6) is given by
Xt =1
r· rx exp
[βWt +
(rK − 1
2β2
)t]
1 + rx∫ t
0exp
[βWs +
(rK − 1
2β2
)s]
ds
=exp
[βWt +
(rK − 1
2β2
)t]
x−1 + r∫ t
0exp
[βWs +
(rK − 1
2β2
)s]
ds.
Example 10.6. Solving the stochastic differential equation
dXt = αt dt + btXt dWt.
Rewrite it as
dXt − btXt dWt = αt dt.
10.1. EXAMPLES AND SOME SOLUTION METHODS 241
����� Example 10.4 ����. ����� ρt such that
ρt dXt − btρtXt dWt = αtρt dt. (10.7)
By integration by parts,
d(ρtXt) = ρt dXt + Xt dρt + d〈ρ,X〉t.
Idea: Suppose Xt dρt = −btρtXt dWt.
(������, ���� idea, ������������.)
Hence, we want to find ρ such that
dρt
ρt
= −bt dWt.
By Ito formula,
−∫ t
0
bu dWu =
∫ t
0
dρu
ρu
= ln(ρt) − ln(ρ0) +1
2
∫ t
0
1
ρ2u
(dρu)2
= ln(ρt) − ln(ρ0) +1
2
∫ t
0
b2u du.
Hence,
ρt = ρ0 exp
(−
∫ t
0
bu dWu − 1
2
∫ t
0
b2u du
).
We may take ρ0 = 1. Thus,
d(ρtXt) = ρt dXt + Xt dρt + (dXt)(dρt)
= ρt dXt − btρtXt dWt + (αt dt + btXt dWt)(−btρtXt dWt)
= ρt dXt − btρtXt dWt − b2t ρtXt dt.
Plugging this result into (10.7), we have
d(ρtXt) + b2t ρtXt dt = αtρt dt.
242 10. STOCHASTIC DIFFERENTIAL EQUATIONS
Using integrating factor again, let Gt = exp
(∫ t
0
b2u du
), we get
d(GtρtXt) = Gt d(ρtXt) + ρtXt dGt = αtρtGt dt.
Set
F (t) = ρtGt = exp
(−
∫ t
0
bu dWu +1
2
∫ t
0
b2u du
).
Thus,
d(FtXt) = αtFt dt,
i.e.,
FtXt − F0X0 =
∫ t
0
αuFu du.
Hence, its solution is given by
Xt = F−1t X0 + F−1
t
∫ t
0
αuFu du
= X0 exp
(∫ t
0
bu dWu − 1
2
∫ t
0
b2u du
)+
∫ t
0
αu exp
(∫ t
u
bv dWv − 1
2
∫ t
u
b2v dv
)du.
Example 10.7. Solving the stochastic differential equation
LQ′′t + RQ′
t +1
CQt = Gt + αWt, (10.8)
where Wt is the white noise. Introduce the vector
Xt =
⎛⎜⎝ X
(1)t
X(2)t
⎞⎟⎠ =
⎛⎜⎝ Qt
Q′t
⎞⎟⎠ ,
then ⎧⎪⎨⎪⎩
(X(1)t )′ = X
(2)t ,
L(X(2)t )′ + RX
(2)t +
1
CX
(1)t = Gt + αWt.
Thus, we may rewrite (10.8) as
dXt = AXt dt + Ht dt + K dWt,
10.2. AN EXISTENCE AND UNIQUENESS RESULT 243
where (Wt) is a 1-dimensional Brownian motion,
Xt =
⎛⎜⎝ X
(1)t
X(2)t
⎞⎟⎠
A =
⎛⎜⎝ 0 1
− 1
CL−R
L
⎞⎟⎠
Ht =
⎛⎜⎝ 0
Gt
L
⎞⎟⎠
K =
⎛⎜⎝ 0
α
L
⎞⎟⎠ .
Thus,
d exp [exp(−At)Xt] = exp(−At)(Ht dt + K dWt).
The solution is of the form
Xt = exp(At)X0 + exp(At)
∫ t
0
exp(−As)(Hs ds + K dWs).
������ coordinate ��.
10.2. An existence and uniqueness result
Theorem 10.8 (Existence and uniqueness theorem for stochastic differential equa-
tion). Let T > 0 and let b, σ (b : [0, T ]×Rn → Rn, σ : [0, T ]×Rn → Rn×m) be measurable
functions satisfying
|b(t, x)| + |σ(t, x)| ≤ C(1 + |x|) for all x ∈ Rn, t ∈ [0, T ] (10.9)
244 10. STOCHASTIC DIFFERENTIAL EQUATIONS
for some constant C, where |σ|2 =n∑
i=1
m∑j=1
|σij|2, and
|b(t, x) − b(t, y)| + |σ(t, x) − σ(t, y)| ≤ D|x − y| for all x, y ∈ Rn, t ∈ [0, T ] (10.10)
for some constant D. Let Z be a random variable which is independent of FWT , the σ-
algebra generated by (Ws : 0 ≤ s ≤ T ), and E|Z|2 < ∞. Then the stochastic differential
equation
dXt = b(t,Xt) dt + σ(t,Xt) dWt, 0 ≤ t ≤ T, X0 = Z
has a unique t-continuous solution Xt with the properties that
(i) Xt is adapted to FWt ∨ σ(Z);
(ii) E[∫ T
0
|Xt|2 dt
]< ∞.
Remark 10.9. The conditions (10.9) and (10.10) are natural in view of the following
two simple examples from deterministic differential equations.
(a) Consider
dXt
dt= X2
t , X0 = 1, t ∈ [0,∞)
corresponding to b(t, x) = x2 (not satisfying (10.9)). This differential equation
has the (unique) solution
Xt =1
1 − t, 0 ≤ t < 1.
But no global solution in this case. More generally, (10.9) ensures that the
solution (Xt) does not explode, i.e., |Xt| does not tend to ∞ in a finite time.
(b) Consider the differential equation
dXt
dt= 3X
23t , X0 = 0 (10.11)
10.3. WEAK AND STRONG SOLUTIONS 245
which has more than one solution. For any a > 0, the function
Xt =
⎧⎪⎨⎪⎩
0 for t ≤ a
(t − a)3 for t > a
is the solution of (10.11).
In this case, b(x) = 3x23 does not satisfy the condition (10.10) at x = 0.
10.3. Weak and strong solutions
Definition 10.10. (1) A strong solution of the stochastic differential equation
dXt = b(t,Xt) dt + σ(t,Xt) dWt (10.12)
on a given probability space (Ω,F , P) and with respect to the fixed Brownian
motion motion W and initial value Z, is a stochastic process X with continuous
sample paths and with the following properties:
(i) X is adapted to the filtration (Ft);
(ii) P[X0 = Z] = 1;
(iii) P[∫ t
0
(|bi(s, xs)| + σ2ij(s,Xs)) ds < ∞
]= 1 for all 1 ≤ i ≤ d, 1 ≤ j ≤ r, and
0 ≤ t < ∞;
(iv) the integral version of (10.12):
Xt = X0 +
∫ t
0
b(s,Xs) ds +
∫ t
0
σ(s,Xs) dWs, 0 ≤ t < ∞.
(2) A weak solution of (10.12) is a triple (X, W ), (Ω,F , P), (Ft), where
(i) (Ω,F , P) is a probability space, and (Ft) is a filtration of sub-σ-algebra of
F satisfying the usual conditions.
246 10. STOCHASTIC DIFFERENTIAL EQUATIONS
(ii) X = (Xt,Ft)0≤t<∞ is a continuous, adapted Rn-valued process. W =
(Wt,Ft)0≤t<∞ is a standard Brownian motion, and (iii),(iv) of (1) are satis-
fied.
����, ��������� strong solution ����� weak solution.
Remark 10.11. There are stochastic differential equations which has no strong solu-
tion, but still has a weak solution.
Example 10.12 (Tanaka equation). Consider the stochastic differential equation
dXt = sign(Xt) dWt (10.13)
where
sign(x) =
⎧⎪⎨⎪⎩
1, if x ≥ 0
−1, if x < 0.
Note that σ(t, x) = sign(x) does not satisfy the Lipschitz condition (10.10). This implies
that we cannot apply Theorem 10.8.
(1) Claim that (10.13) has no strong solution.
Suppose X is a strong solution of (10.13). Then X is a Brownian motion by Levy
Theorem. Moreover,
dWt = sign(Xt) dXt.
This implies that the stochastic process (Wt) given by
Wt =
∫ t
0
sign(Xu) dXu
is a Brownian motion with respect to (FXt ). By the Tanaka equation ,
Wt = |Xt| − 2Lxt ,
10.4. FEYNMAN-KAC FORMULA 247
where Lxt is the local time of (Xt). This means that FW
t � FXt . This contradicts
to X is a strong solution.
(2) Find a weak solution of (10.13)
Choose (Xt) to be a Brownian motion (Bt). Then we can define (Bt) by
Bt :=
∫ t
0
sign(Bu) dBu =
∫ t
0
sign(Xu) dXu,
i.e., dBt = sign(Xt) dXt. Thus,
dXt = sign(Xt) dBt.
Hence, (Xt) is a weak solution.
10.4. Feynman-Kac formula
����������������.
Theorem 10.13 (Feynman-Kac formula). Consider the stochastic differential equation
dXt = β(t,Xt) dt + γ(t,Xt) dWt.
Let f be a Borel-measurable function. Fix T > 0 and let t ∈ [0, T ] be given. Define the
function
g(t, x) = Et,x[f(XT )] = E[f(XT )|Xt = x].
Assume that g(t, x) < ∞ for all (t, x). Then g(t, x) satisfies the partial differential equa-
tion
gt(t, x) + β(t, x)gx(t, x) +1
2γ2(t, x)gxx(t, x) = 0
with the terminal condition g(T, x) = f(x) for all x.
248 10. STOCHASTIC DIFFERENTIAL EQUATIONS
Remark 10.14. (g(t, Xt))0≤t≤T is a martingale.
Theorem 10.15 (Discounted Feynman-Kac formula). Consider the stochastic differ-
ential equation
dSt = rβ(t, St) dt + σγ(t, St) dWt
Let f be a Borel-measurable function and let r be constant. Fix T > 0 and let t ∈ [0, T ]
be given. Define the function
h(t, x) = Et,x[e−r(T−t)f(XT )] = E[e−r(T−t)f(XT )|Xt = x]
Assume that h(t, x) < ∞ for all (t, x). Then h(t, x) satisfies the partial differential equa-
tion
ht(t, x) + β(t, x)hx(t, x) +1
2γ2(t, x)hxx(t, x) = rh(t, x)
with the terminal condition h(T, x) = f(x) for all x.
� stochastic analysis ��� Feynmann-Kac Theorem �������. ����
����, ��������� finance �����.
Example 10.16. Suppose that (St) satisfies the stochastic differential equation
dSt = rSt dt + σSt dWt.
Let
f(t, x) = Et,x[h(St)].
Then f(t, x) satisfies the partial differential equation
ft(t, x) + rxfx(t, x) +1
2σ2x2fxx(t, x) = 0.
with terminal condition h(T, x) = f(x). This equation can be solved numerically.
10.4. FEYNMAN-KAC FORMULA 249
Theorem 10.17 (2-dimensional Feynman-Kac formula). Let Wt = (W 1t ,W 2
t ) be a
2-dimensional Brownian motion. Consider two stochastic differntial equations
dXt = β1(t,Xt, Yt) dt + γ11(t,Xt, Yt) dW 1t + γ12(t,Xt, Yt) dW 2
t ,
dYt = β2(t,Xt, Yt) dt + γ21(t,Xt, Yt) dW)t + γ22(t,Xt, Yt) dW 2
t .
Let f(x, y) be a Borel measurable function, define
g(t, x, y) = Et,x,y[f(XT , YT )],
h(t, x, y) = Et,x,y[e−r(T−t)f(XT , YT )].
Then g and h satisfy the partial differential equations
gt + β1gx + β2gy +1
2(γ2
11 + γ212)gxx + (γ11γ21 + γ12γ22)gxy +
1
2(γ2
21 + γ222)gyy = 0,
ht + β1hx + β2hy +1
2(γ2
11 + γ212)hxx + (γ11γ21 + γ12γ22)hxy +
1
2(γ2
21 + γ222)hyy = rh
with terminal conditions
g(T, x, y) = h(T, x, y) = f(x, y), for all x, y.