Solutions of the homework #2

LU ANG
April 24th, 2020

Question 1
(1)
According to previous study we know that the β̂OLS is derived as:

PT PT n
t=1 (Xt − X̄)(yt − ȳ) (Xt − X̄)yt X
β̂OLS = PT = Pt=1
T
= ~ · ~y
ωt yt = ω
2 2
t=1 (Xt − X̄) t=1 (Xt − X̄) t=1

Xt − X̄
where ωt = PT
t=1 (Xt − X̄)2
if we view β̂OLS = f (~y ) , we can easily prove that:

f (α~y ) = αf (~y ) and f (~y + ~z) = f (~y ) + f (~z)

which means that β̂OLS is a linear estimator

1
[Econometrics] Lu Ang

(2)
β̂OLS is unbiased estimator means E(β̂OLS ) = β , first recall that:
T
X T
X
β̂OLS = ωt yt = ωt (α + βXt + ut )
t=1 t=1
XT Xn T
X
=α ωt + β ωt Xt + ωt ut
t=1 t=1 t=1
T
X
=β+ ωt ut
t=1

then we take expectation of β̂OLS :

T
X T
X
E(β̂OLS ) = E(β + ωt ut ) = β + ωt E(ut ) = β q.e.d
t=1 t=1

(3)
Suppose that β̃ is any unbiased linear estimator, which can be written as:
T
X T
X
β̃ = ct y t = (ωt + dt )yt
t=1 t=1
XT
= (ωt + dt )(α + βXt + ut )
t=1
XT T
X T
X T
X T
X T
X
=α ωt +β ωt Xt + ωt ut + α dt + β dt Xt + dt u t
t=1 t=1 t=1 t=1 t=1 t=1
| {z } | {z }
=0 =1
T
X T
X T
X T
X
=β+α dt + β dt Xt + ωt ut + dt u t
t=1 t=1 t=1 t=1

Then we take expectation of β̃, notice that E(ut ) = 0

T
X T
X T
X T
X
E(β̃) = β + α dt + β dt xt + ωt E(ut ) + dt E(ut )
| {z } | {z }
t=1 t=1 t=1 =0 t=1 =0

2
[Econometrics] Lu Ang

T
X T
X
E(β̃) = β + α dt + β d t xt = β
t=1 t=1
T
X T
X
i.e. dt = 0 and d t xt = 0 (1)
t=1 t=1

Next we take the variance of β̃, notice that V (ut ) = σ 2 .

T
X T
X T
X
V (β̃) = V (β + (ωt + dt )ut ) = V ( (ωt + dt )ut ) = V ((ωt + dt )ut )
t=1 t=1 t=1
T
X
= (ωt + dt )2 V (ut )
t=1
XT T
X T
X
= σ2( ωt2 + ωt dt + d2t )
t=1 t=1 t=1

According to result (1) we know that:

=0 =0
z }| { z }| {
XT XT

T PT dt Xt −X̄ dt
X (Xt − X̄)dt t=1 t=1
ωt dt = Pt=1
T
= PT =0
t=1 t=1 (Xt − X̄)2 t=1 (Xt − X̄)
2

then we can rewrite V (β̃) as:

T
X T
X
V (β̃) = σ 2 ( ωt2 + d2t )
t=1 t=1
T
X T
X
= σ2 ωt2 + σ 2 d2t
t=1 t=1
XT
= V (β̂OLS ) + σ 2 d2t ≥ V (β̂OLS )
t=1

Thus we have proved the efficiency of OLS estimator

3
[Econometrics] Lu Ang

(4)
according to weak law of large number(WLLN):
T
1X p
(Xt − X̄)(ut − ū) → Cov(Xt , ut ) = 0
T t=1

P.S :

Cov(Xt , ut ) = E(Xt ut )−E(Xt ) E(ut ) = E(E(Xt ut |Xt )) = E(Xt E(ut |Xt )) = 0

| {z } | {z }
=0 =0

we also assume that:

T
1X p
(Xt − X̄)2 → m < ∞
T t=1

Then we look at β̂OLS :

T 1
PT
t=1 (Xt − X̄)(ut − ū)
X
T
β̂OLS = β + ωt ut = β + 1
PT 2
t=1 T t=1 (Xt − X̄)

1
PT
T t=1 (Xt − X̄)(ut − ū) 0
as T −→ ∞, 1
PT −→
T t=1 (Xt − X̄)
2 m
thus we have:
β̂OLS −→ β as n −→ ∞ q.e.d

4
(5)
Following the Center Limit theorem(CLT):

=0
PT PT PT PT z }| {
ω u
t=1 t t − E( ω
t=1 t t u ) ω u
t t − ωt E(u t)
q P = t=1 q P t=1
V ( Tt=1 ωt ut ) V ( Tt=1 ωt ut )
PT
ωt ut
= qt=1
PT 2
σ t=1 ωt

β̂ − β
= qP
T 2
σ/ t=1 (Xt − X̄)
√
T β̂ − β
= q P −→ N (0, 1)
1 T 2
σ/ T t=1 (Xt − X̄)

PT PT 1
where t=1 ωt ut = β̂ − β and t=1 ωt2 = PT
t=1 (Xt − X̄)2

1
PT
as T −→ ∞ we substitute T t=1 (Xt − X̄) with its convergence value m:

√
T (β̂ − β)
√ −→ N (0, 1)
σ/ m

√ σ2
T (β̂ − β) −→ N (0, )
m