Important results
Covariance properties​
Cov(X,a)Cov(X,X)Cov(X,Y)Cov(aX,bY)Cov(X+a,Y+b)Cov(aX+bY,cW+dV)​=0=Var(X)=Cov(Y,X)=abCov(X,Y)=Cov(X,Y)=acCov(X,W)+adCov(X,V)+bcCov(Y,W)+bdCov(Y,V)​
Mean independence and Covariance​
Prove: If X and U are mean independent and E[U]=0 then Cov(X,U)=0.
Note: Mean independence is defined as E[U∣X]=E[U].
Proof:
Cov(X,U)​=E[(X−E[X])(U−E[U])]=E[XU−XE[U]−E[X]U+E[X]E[U]]=E[XU]−E[X]E[U]−E[X]E[U]+E[X]E[U]=E[XU]−E[X]E[U]=E[XU]−E[X]E[E[U∣X]]=E[XU]−E[X]E[0]=E[XU]=E[E[XU∣X]]=E[XE[U∣X]]=E[X⋅0]=0​
Unbiased estimator of Variance​
Let the population variance and mean are σ2 and μ. The sample variance is given by:
S2=n−11​i=1∑n​(Xi​−Xˉ)2
To prove the unbiasedness of S2 we have to prove that E[S2]=σ2.
S2=n−11​i=1∑n​(Xi2​+Xˉ2−2Xi​Xˉ)
Taking expectation both sides:
E[S2]​=E[n−11​i=1∑n​(Xi2​+Xˉ2−2Xi​Xˉ)]=n−11​E[i=1∑n​(Xi2​)+i=1∑n​(Xˉ2)−2Xˉi=1∑n​(Xi​)]=n−11​E[i=1∑n​(Xi2​)+n(Xˉ2)−2Xˉn(Xˉ)]=n−11​E[i=1∑n​(Xi2​)−n(Xˉ2)]=n−11​[i=1∑n​E[Xi2​]−nE[Xˉ2]] as E operator is linear.​
We know that Var[Xi​]=E[Xi2​]−(E[Xi​])2, that is σ2=E[Xi2​]−μ2, hence
n−11​[i=1∑n​E[Xi2​]−nE[Xˉ2]]​=n−11​[i=1∑n​(σ2+μ2)−nE[Xˉ2]]=n−11​[n(σ2+μ2)−nE[Xˉ2]]=n−1n​[(σ2+μ2)−E[Xˉ2]]​
Also, Var[Xˉ]=E[Xˉ2]−(E[Xˉ])2, ⟹E[Xˉ2]=Var[Xˉ]+(E[Xˉ])2.
Now
Var[Xˉ]​=Var[n∑i=1n​Xi​​]=n21​Var[i=1∑n​Xi​]=n21​i=1∑n​Var[Xi​]( as Xi​’s are i.i.d)=n2nσ2​=nσ2​​
- Consider (E[Xˉ])2,
E[Xˉ](E[Xˉ])2​=E[n∑i=1n​Xi​​]=n1​E[i=1∑n​Xi​]=n1​i=1∑n​E[Xi​]=n1​i=1∑n​E[Xi​]=μ=μ2​
Finally, we have
n−1n​[(σ2+μ2)−E[Xˉ2]]E[S2]​=n−1n​[(σ2+μ2)−Var[Xˉ]−(E[Xˉ])2]=n−1n​[(σ2+μ2)−nσ2​−μ2]=n−1n​[n(n−1)σ2​]=σ2■​
Exercise: Find E[S2], where S is given by:
n1​i=1∑n​(Xi​−Xˉ)2
Q is positive definite​
Prove:
If XN×k​ has full column rank, then Q:=N1​(X′X) is a positive definite matrix.
A matrix A is positive definite if for all non-zero vector vk×1​, the quadratic form v′Av is positive.
Proof:
We have to show
v′N(X′X)​v>0.
Above inequality can be written as
N1​v′X′XvN1​(Xv)′Xv​>0>0​
Since (Xv) is vector of dimension N×1, (Xv)′Xv is the inner product of (Xv) with itself, which is always non-negative (≥) by the definition of inner product.
If X has full column rank, then the equation Xv=0 only holds true if v=0. Therefore Xvî€ =0 for vî€ =0.
Hence
N1​v′(X′X)v​>0.■​
Law of Total Probability​