Solution#2
Shouto Yonekura
August 1, 2016
(1)
First we can decompose ˆβ as follows:
β = (Xˆ ′X)−1X′y
= (X′X)−1X′(Xβ + u)
= β + (X′X)−1X′u
= β + (n1X′X)−1 1nX′u
= β + Q−1xxQxu. Using the LLN, we can easily that Qxu→p0 since
1 nX
′u →pX′E[u]
= 0.
Suppose that Qxxconverges to some finite matrix Mxx. Then QxxQxu→p0. Hence we get ˆβ →pβ.
(2) (3)
Suppose that assumptions A1 ∼ A5 hold. In addtion to these, E[u4i] and (xix
′
i)2are finite for any i. Where xi∈ Rn. By the triangle inequality and Jensen’s inequality, we can show that
k E[xix′iu2i] k2≤ E[k xix′iu2i k2]
= E[k x2iu2i k] =k xik2E[u2i].
Using Cauchy–Schwarz inequality, we will obtain
k xik2E[u2i] ≤ (k xik4)1/2(E[u4i])1/2
< ∞.
Therefore, we finally get √n1 Pixiui→dN (0, Ω) by CLT. Next, we can rewriten√n( ˆβ − β) as follows:
√n( ˆβ − β) =√n(X′X)−1X′u,
= (n1X′X)−1√n1 X′u,
= Q−1xx√n1 X′u.
1
From Prop11.11,√n1 X′uui→dN (0, Ω).Suppose that Qxxconverges to some finite matrix Mxx. Then using Slutsky’s theorem, we obtain that Q−1xx√n1 X′u →dMxx−1N (0, Ω). Hence we can finally show that:
√n( ˆβ − β) →dNk(0, Mxx−1ΩMxx−1) , (Mxx′ = Mxx)
= Nk(0, V ) Q.E.D.
(4)
Suppose that Xi∼iid f (xi; θ). Then the likelihood function is given by L(θ) :=Qni=1f (θ; xi)
(5)
First we can show that
l′′(θ; X) = ∂θ(∂θf (θ;X)f (θ;X))
= ∂
2 θ2f (θ;X)
f (θ;X) −∂θf (θ;X)f (θ;X)
2
holds. Multiplying both side by f (θ; X) and integrating with respect to x, we get E[l′′(θ; X)] =´ ∂θ22f (θ;X)f (θ;X)f (θ; X)dx −∂θf (θ;X)f (θ;X)2f (θ; X)dx
=´ ∂θ22f (θ; X)dx − I(θ)
∂θ22´ f (θ; X)dx − I(θ)
= −I(θ).
Therefore, I(θ) = −E[∂θ22lnf (θ; X)] holds. Let T (X) be an unbiased estimator of θ. Next we get
∂θE[T (X)] = ∂θ´ T (X)f (x; θ)dx
⇐⇒ 1 = ∂θ´ T (X)f (x; θ)dx
=´ ∂θT (X)f (x; θ)dx
=´ T (X)∂θlnfX(x; θ)fX(x; θ)dx E[T (X)l′(θ)].
Since E[l′(θ)] = 0 (Prop9.2,) this can be rewriten as follows:
E[T (X)l′(θ)] = E[(T (X) − θ)l′(θ)]
= Cov(T (X), l′(θ)).
2
This leads to
1 = Cov((T (X), l′(θ)))2≤ V [T (X)]V [l′(θ)] , (−1 ≤ Cov(X, Y ) pV [X]pV [Y ] ≤ 1)
= V [T (X)]I(θ). Therefore, V [T (X)] ≥ 1/I(θ) holds. Q.E.D.
(6)
Assume that parameter space Θ is a compact set. Let θ0 be a true value of parameter θ. Then for any θ06= θ, we
have by Jensen’s inequality
Eθ0
hlnf (x;θf (x;θ)
0)
i
≤ lnEθ0
hlnf (x;θf (x;θ)
0)
i= 0,
since
Eθ0
hf (x;θ)
f (x;θ0)
i=´ f (x;θf (x;θ)0)f (x; θ0)dx
=´ f (x; θ)dx
= 1,
in the case of continious X. (with an analogous story in the case of discrete X). Fix δ > 0. Let a1:= Eθ0
hlnf (x;θf (x;θ0−δ)0) i< 0, a2:= Eθ0
hlnf (x;θf (x;θ0+δ)0) i< 0. Usinng LLN and continuity of l(θ) := lnL(θ), we get
l(θ0−δ)−l(θ0) n →pa1,
thus l(θ0− δ) < l(θ0) for any sufficiently large n. Similaly, we have l(θ0+ δ) < l(θ0) for any sufficiently large n. Since l(θ) is a continious function with respect to θ and Θ is a compact set, there exists ˆθ such that miximizes l(θ) on the interval (θ0− δ, θ0+ δ). This menas that ˆθ is a consistent estimator.
Next we assume that l(θ) is twice continiously defferentiable on a negihrhood of θ0. We have already shown that there exixts ˆθ such that l′(ˆθ) = 0 and ˆθ →pθ0. A Taylor expansion gives that
−l′(θ0) = l′(ˆθ) − l′(θ0) = (ˆθ − θ0)l””(θ#), where θ#lies on between θ0 and ˆθ. Therefore
(ˆθ − θ0) = −l””l′(θ(θ0#)),
3
and we can rewrite this as follows:
pnI(θ0)(ˆθ − θ0) = l
′(θ0)
√nI(θ0) l′′(θ0) l′′(θ#)
−l
′′(θ0) nI(θ0)
−1
.
First, usging CLT, we get
l′(θ0)
√nI(θ0) →d N (0, 1).
Next, since l(θ) is twice continiously defferentiable on a negihrhood of θ0 , LLN and continious mapping theorem yield
n−1l′′(θ0) n−1l′′(θ#)→p
E[l′′(θ0)] E[l′′(θ0)]
= −I (θ0)
−I (θ0)= 1.
Finally, LLN also gives
n−1l′′(θ0) →p−I(θ0). Thus
−l
′′(θ0) nI(θ0)
→p1. From these results, we can conclude that
√n(ˆθ − θ0) → N(0, I(θ0)−1).
Q.E.D.
4