1. Martingales (鞅)
1.1 Histories of probability space
(
Ω
,
F
,
P
)
(\Omega,\mathcal{F},P)
(Ω,F,P) probability space
(
X
t
,
t
≥
0
)
(X_t, t\ge 0)
(Xt,t≥0) stochastic process
定义. 随机变量生产的
σ
\sigma
σ-代数:In association with such a process
X
t
X_t
Xt, we define for each
t
≥
0
t\ge 0
t≥0 a sub-
σ
\sigma
σ-filed of
F
\mathcal{F}
F, denoted
F
t
X
\mathcal{F}_t^{X}
FtX, by
F
t
X
=
σ
(
X
s
,
s
∈
[
0
,
t
]
)
\mathcal{F}_t^{X} = \sigma(X_s, s\in[0,t])
FtX=σ(Xs,s∈[0,t])
举例:随机变量生成的
σ
\sigma
σ-代数,指的是一组特殊事件组成的集合:这些事件是否发生,可以通过随机变量的取值明确判断出来。举个例子吧。
比如今天可能下雨也可能不下,下雨时随机变量
X
=
1
X=1
X=1,反之
X
=
0
X=0
X=0。然而,下不下雨只是“今天”的一个属性,其它属性,比如我早饭吃的是火腿还是培根,也是全世界所包含的信息的一部分。
假设整个概率空间由下面这些元素组成:
[火腿,下雨]:概率0.25,X = 1
[火腿,不下雨]:概率0.25,X = 0
[培根,下雨]:概率0.25,X = 1
[培根,不下雨]:概率0.25,X = 0
这时,X生成的代数包括下面三个非空集合:
{X = 1} = {[火腿,下雨],[培根,下雨]}
{X = 0} = {[火腿,不下雨],[培根,不下雨]}
{X = 0或1} = {[火腿,不下雨],[培根,不下雨],[火腿,下雨],[培根,下雨]}
可见不管X=1还是0,我都既有可能吃火腿,也有可能吃培根;从X的取值里,你得不到任何关于我早饭吃了什么的信息。因此“早饭吃培根”这一事件,就被排除在X生成的代数之外了。具体来说:早饭吃培根 = {{培根,下雨},{培根,不下雨}},不属于X生成的代数。
定义. Let
Y
Y
Y be a
R
k
\mathbb{R}^k
Rk- valued process such that for each
t
≥
0
t\ge 0
t≥0,
Y
t
Y_t
Yt is
F
t
X
\mathcal{F}_t^X
FtX-measurable, a well-known representation result on measurability states that
Y
=
ϕ
(
X
s
,
s
∈
S
)
Y=\phi(X_s,s\in S)
Y=ϕ(Xs,s∈S) where
S
S
S is a countable subset of
[
0
,
t
]
[0,t]
[0,t] and
ϕ
\phi
ϕ is a borelian function(Borel函数) from
R
n
∣
S
∣
\mathbb{R}^{n|S|}
Rn∣S∣ into
R
k
\mathbb{R}^k
Rk.
We say
Y
t
Y_t
Yt is adapted to
F
t
X
\mathcal{F}_t^X
FtX, if
Y
t
=
ψ
t
(
X
0
t
)
Y_t=\psi_t(X_0^t)
Yt=ψt(X0t) for some
ψ
t
\psi_t
ψt, i.e.,
Y
t
Y_t
Yt depends causally on
X
t
X_t
Xt.
If F t ⊇ F t X \mathcal{F}_t\supseteq\mathcal{F}_t^X Ft⊇FtX, then F t \mathcal{F}_t Ft is called a history of X t X_t Xt, and X t X_t Xt is adapted to F t \mathcal{F}_t Ft.
定义. 停时(stopping time) 一类随机时刻,指具有某种与将来无关性质的随机时刻。给定概率空间 ( Σ , F , P ) (\Sigma,\mathcal{F},P) (Σ,F,P)及其滤子 F t \mathcal{F}_t Ft,映射 τ : Σ → T ∪ { ∞ } \tau:\Sigma→T\cup\{\infty\} τ:Σ→T∪{∞},如果对任意的 t ∈ I t∈I t∈I, { ω : ω ∈ Σ , τ ( ω ) ≤ t } ∈ F t \{\omega:\omega\in\Sigma,τ(ω)≤t\}\in\mathcal{F}_t {ω:ω∈Σ,τ(ω)≤t}∈Ft,则称映射 τ \tau τ为一个 F t \mathcal{F}_t Ft停时。
1.2 Martingale
A history is given on the probability space
(
Ω
,
F
,
P
)
(\Omega,\mathcal{F},P)
(Ω,F,P). A
(
P
,
F
t
)
(P,\mathcal{F}_t)
(P,Ft)-martingale over
[
0
,
c
]
[0,c]
[0,c] is a real-valued stochastic process
X
t
X_t
Xt such that:
(1)
X
t
X_t
Xt is adapted to
F
t
\mathcal{F}_t
Ft,
(2)
X
t
X_t
Xt is
P
P
P-integrable, i.e.,
E
[
∣
X
t
∣
]
<
∞
,
∀
t
∈
[
0
,
c
]
E[|X_t|]<\infty,\forall t\in[0,c]
E[∣Xt∣]<∞,∀t∈[0,c]
(3) for all
0
≤
s
≤
t
≤
c
0\leq s\leq t\leq c
0≤s≤t≤c,
E
[
X
t
∣
F
s
]
=
X
s
E[X_t|\mathcal{F}_s]=X_s
E[Xt∣Fs]=Xs,
P
−
a
.
s
P-a.s
P−a.s (almost surely).
1.3 Levy Formula
Let
X
t
X_t
Xt be a right-continuous
N
+
N_+
N+-valued
F
t
\mathcal{F}_t
Ft-Markov chain which is stable and conservative and admits the
Q
Q
Q-matrix
(
q
i
j
:
i
,
j
∈
N
+
)
(q_{ij}:i,j\in N_+)
(qij:i,j∈N+). If
f
f
f is a nonnegative function from
N
+
×
N
+
N_+\times N_+
N+×N+ into
R
+
R_+
R+, then for any
0
≤
s
≤
t
0\leq s\leq t
0≤s≤t, the Levy formulat holds:
E
[
∑
s
<
u
≤
t
f
(
X
u
−
,
X
u
)
∣
σ
(
X
s
)
]
=
E
[
∫
s
t
∑
j
≠
X
u
q
X
u
j
f
(
X
u
,
j
)
d
u
∣
σ
(
X
s
)
]
E\left[ \mathop{\sum}\limits_{s<u\leq t} f(X_{u^-},X_u) \Big| \sigma(X_s) \right] = E\left[ \int_s^t \mathop{\sum}\limits_{j\neq X_u} q_{X_uj} f(X_u, j) du \Big| \sigma(X_s) \right]
E[s<u≤t∑f(Xu−,Xu)∣∣∣σ(Xs)]=E⎣⎡∫stj=Xu∑qXujf(Xu,j)du∣∣∣σ(Xs)⎦⎤
1.4 Radon-Nikodyn Derivatives (likeihood ratios)
Let
P
P
P and
Q
Q
Q be two probability measures defined on the same measurable space
(
Σ
,
F
)
(\Sigma,\mathcal{F})
(Σ,F), and let
F
t
\mathcal{F}_t
Ft be a history. For each
t
≥
0
t\ge 0
t≥0, denote by
P
t
P_t
Pt and
Q
t
Q_t
Qt the restrictions of
P
P
P and
Q
Q
Q to
F
t
\mathcal{F}_t
Ft respectively. Suppose that for some
c
≥
0
c\ge 0
c≥0,
Q
c
Q_c
Qc is absolutely continuous with respect to
P
c
P_c
Pc. Then clearly, for all
t
∈
[
0
,
c
]
t\in[0,c]
t∈[0,c],
Q
t
Q_t
Qt is absolutely continuous with respect to
P
t
P_t
Pt. Define
L
t
:
=
d
Q
t
d
P
t
L_t := \frac{dQ_t}{dP_t}
Lt:=dPtdQt to be the Radon-Nikodym derivation of
Q
t
Q_t
Qt with respect to
P
t
P_t
Pt.
Proposition. L t L_t Lt is a ( P , F t ) (P,\mathcal{F}_t) (P,Ft)-martingale over [ 0 , c ] [0,c] [0,c].
1.5 Predictability
Exercise. Given a stopping time
τ
\tau
τ, let
I
n
=
{
1
,
if
n
≤
τ
0
,
if
n
>
τ
I_n = \begin{cases} 1, & \text{if }n\leq\tau \\ 0, & \text{if }n>\tau \end{cases}
In={1,0,if n≤τif n>τ Show that
(
I
n
)
n
≥
1
(I_n)_{n\ge 1}
(In)n≥1 is a predictable process.
2. Girsanov Theorem
定理1. If
P
P
P and
Q
Q
Q are equivalent measures, and
X
t
X_t
Xt is an
F
t
\mathcal{F}_t
Ft-adpated process then the following results hold:
E
Q
(
X
t
)
=
E
P
(
X
t
d
Q
d
P
)
E_Q(X_t)=E_P\left(X_t\frac{dQ}{dP}\right)
EQ(Xt)=EP(XtdPdQ)