跳至主要内容

動差(Moments)

期望值和方差

期望值的性質:

  1. E[c]=cE[c]=c for any constant cc.

  2. E[ag(x)+b]=aE[g(X)]+bE[ag(x)+b]=aE[g(X)]+b

  3. g(x)0,x    E[g(X)]0g(x)\ge 0, \forall x \implies E[g(X)]\ge 0

    hence XY    E[X]E[Y]X\ge Y \implies E[X]\ge E[Y]

  4. E[g(X)]E[g(X)]|E[g(X)]|\le E[|g(X)|]

  5. r>0,E[g(X)r]<    E[g(X)s]<,0<srr>0, E[|g(X)|^r]<\infty \implies E[|g(X)|^s]<\infty, \forall 0<s\le r

  6. XY    g(X)h(Y)X\perp Y\implies g(X)\perp h(Y), and E[g(X)h(Y)]=E[g(X)]E[h(Y)]E[g(X)h(Y)]=E[g(X)]E[h(Y)]

Let XX be a r.v. with E[X]E[X] exists, then c\forall c

E[(Xc)2]=E[XE[X]+E[X]c]2=E[(XE[X])2]+2E[(XE[X])(E[X]c)]+(E[X]c)2=E[(XE[X])2]+(E[X]c)2\begin{align*} E[(X-c)^2] &= E[X-E[X]+E[X]-c]^2\\ &=E[(X-E[X])^2] + 2E[(X-E[X])(E[X]-c)] + (E[X]-c)^2\\ &=E[(X-E[X])^2] + (E[X]-c)^2 \end{align*}

i.e. E[(Xc)2]E[(XE[X])2]E[(X-c)^2]\ge E[(X-E[X])^2], and the equality holds if and only if c=E[X]c=E[X].

也就是说 E[X]E[X]XX 的均值(中心),因为每个点到 E[X]E[X] 的距离的平方的期望值最小。

Definition

Variance

XX 每个點 到中心的距离的平方的期望值,即 XX 的分散程度,称为 XX 的方差。

σ2(X)=alsoVar(X)=E[(XE[X])2]=E[X2](E[X])20\sigma^2(X)\xlongequal{also} Var(X) = E[(X-E[X])^2]=E[X^2]-(E[X])^2\ge 0

方差的性質:

  1. Var(c)=0Var(c)=0

  2. Var(ag(X)+b)=a2Var(g(X))Var(ag(X)+b)=a^2Var(g(X))

  3. Var(X+Y)=E[(X+YE[X]E[Y])2]=E[(XE[X])2]+E[(YE[Y])2]+2E[(XE[X])(YE[Y])]=Var(X)+Var(Y)+2Cov(X,Y)\begin{align*} Var(X+Y) &= E[(X+Y-E[X]-E[Y])^2]\\ &= E[(X-E[X])^2] + E[(Y-E[Y])^2] + 2E[(X-E[X])(Y-E[Y])]\\ &= Var(X) + Var(Y) + 2Cov(X,Y) \end{align*}

    with Cov(X,Y)E[(XE[X])(YE[Y])]=E[XY]E[X]E[Y]Cov(X,Y)\triangleq E[(X-E[X])(Y-E[Y])]=E[XY]-E[X]E[Y] is called the covariance(共變數) of XX and YY.

也就是說,當 XE[X]X-E[X]YE[Y]Y-E[Y] 同號時, Cov(X,Y)>0Cov(X,Y)>0,異號時 Cov(X,Y)<0Cov(X,Y)<0

CovCov 進行歸一化(unit-free)得到

Cov(X,Y)Var(X)Var(Y)ρX,Y\frac{Cov(X,Y)}{\sqrt{Var(X)Var(Y)}}\triangleq \rho_{X,Y}

稱為 XXYY 的相關係數(correlation coefficient)。

如果 XY    E[XY]=E[X]E[Y]    Cov(X,Y)=0    ρX,Y=0X \perp Y \implies E[XY]=E[X]E[Y]\implies Cov(X,Y)=0\implies\rho_{X,Y}=0 and Var(X+Y)=Var(X)+Var(Y)Var(X+Y)=Var(X)+Var(Y)

動差(Moments)

Definition

The Moment Generation Function of XX is defined as

MX(t)E[etX],tRM_X(t)\triangleq E[e^{tX}],\quad t\in\mathbb{R}

Facts:

  1. MX(0)=E[E0X]=E[1]=1M_X(0)=E[E^{0X}]=E[1]=1
  2. If MX(t)M_X(t) exists t<δ\forall |t|<\delta for some δ>0\delta>0, then E[Xk]E[X^k] exists k=1,2,\forall k=1,2,\cdots and is given by E[Xk=MX(k)]t=0=MX(t)(0)E[X^k=M_X^{(k)}]|_{t=0}=M_X^{(t)}(0)
  3. X=dY    MX(t)=MY(t)X\xlongequal{d}Y \iff M_X(t)=M_Y(t)