Large-N and Large-T Properties of Panel Data Estimators and the Hausman Test



Let Xh,2,it be the hth element of Xh,2,it. The required result follows if we can
show that

sup
i,T


T X (xh,2,it
T t


- Exh,2,it)


< M for all h.

4


(41)


The proof of (41) is similar to the proof of Lemma 1 of Andrews (1991). This
proof relies on the following
α- mixing inequality presented in Hall and Heyde
(1980, p. 278). Suppose that
Y and W are random variables that are G-
measurable and H- measurable, respectively, with EY p and EWq <
, where p,q > 1 with 1/p + 1/q < 1. Then,

E (Y - EY ) (W - EW ) ∣ ≤ 8 H Y ∣∣p ∣∣ W ∣∣g [α (G, H)]1-1/p-1/q ,     (42)

where α (G, H) is the α-mixing coefficient between the sigma fields G and H.
Now, let
Xit = xh,2,it - Exh,2,it. Notice that

4

p e .. Σ X«)

sup T χ x x xe (Xi,XisXipXik )

i,τ      t=1 s = 1 p=1 k= 1

T -tT-sT -p

⅛ 5⅛2χχχΣE (XitXi,t+sXi,t+s+pXi,t+s+p+k)
i,T      t s=0 p=0 k=0

4!siuτp T X


Σ

0p,ks
0p+k+sT-t


E (Xit (Xi,t+sXi,t+s+pXi,t+s+p+k))


+4!sup ɪ X X

i,T        t     0s,kp


0p+k+sT-t


+4!sup .=2 X X

i,i        t     0s,kp

0p+k+sT-t


1

+4!siup .χ


x

0s,pk
0p+k+sT-t


= I + II + III + IV, say.


E [(Xit

Xi,t+s) (Xi,t+s+pXi,t+s+p+k)]

-E (XitXi,t+s) E (Xi,t+s+pXi,t+s+p+k )


E (XitXi,t+s) E (Xi,t+s+pXi,t+s+p+k)


E ((Xit

Xi,t+sXi,t+s+p') Xi,t+s+p+k)


By applying the inequality of (42) to Xi,t+sXi,t+s+pXi,t+s+p+k and Xit and
then by the Holder inequality, we have

1T

i4!8sup .2                    HXitk4q HXi,t+sH4q Il Xi,t+s+p ∣∣ 4q

i,T     t=10p,ksT-t

×∣∣Xi,t+s+p+k∣∣4q αi (s)q-1


35




More intriguing information

1. From music student to professional: the process of transition
2. Informal Labour and Credit Markets: A Survey.
3. Land Police in Mozambique: Future Perspectives
4. The name is absent
5. Who is missing from higher education?
6. The name is absent
7. Workforce or Workfare?
8. The name is absent
9. Convergence in TFP among Italian Regions - Panel Unit Roots with Heterogeneity and Cross Sectional Dependence
10. The name is absent