A Principal Components Approach to Cross-Section Dependence in Panels



Suppose data are available on xit and xit and we focus on the regressions

yit = θxit + uit         (RI)

(7)


yit = bxit + cwt + vit (RII)

which differ in that the latter is augmented by wt = PiN=1 uit/N to proxy the
unobserved random factor
zt. It can be shown that, for our baseline DGP, this
proxy is equal to the
first principal component of u up to a scaling factor. To
study the properties of these regressions, we de
fine the sequential probability
limits (
plim) for T →∞and any fixed N

SyNy =plimT-1   tyi2t; SyNx =plimT-1 t yitxit           (8)

T→∞              T→∞

and likewise for the other variables in (6) whose (co)variances are:

E (xi2t) = σ2x = σd2 + σz2

E(di2t) = σ2d
E(εi2t)=σε2
E(xitxjt) = σz2
E(xitdit) = σ2d
E(xitzt) = σz2
E(zt2) = σz2


E(yi2t) = β2σ2d +(β + γ)2σz2 +σε2

E(xityit) = βσd2 +(β + γ)σz2

E(yitzt)=(β+γ)σz2

E(yityjt)=(β + γ)2σz2

E(ztdit) = E(ztεit) = E(ditεit) = 0

The auxiliary regression zt = δxit +ηit implies that yit = (β+γδ)xit+γηit+εit .
Hence, the OLS estimator θ measures (β + γδ) and the residuals Uit estimate
uit = γηit + εit, with variance σ2u E(ui2t)=γ2 (1 - δ)2σz2 + γ2 δ2 σ2d + σε2 and
covariance
σij E(uitujt) = γ2 (1 - δ)2σz2. Then we have wt = PiN=1(γηit +
εit)/N = γ(1 — δ)zt γδdt + εt where dt = N-1 ɪɪi dit and it follows that

E(wt2) = γ2(1 - δ)2σz2 + γ2δ2N-1σ2d + N-1σε2
E(yitwt)=(β+γ)(1 — δ)γσz2 βγδN-1σ2d + N-1σε2
E(xitwt)  =  γ(1 — δ)σz2 γδN-1σ2d;  E(ztwt) = γ(1 — δ)σz2

For our baseline DGP we have

SyNy = β2σd2 + (β + γ)2σz2 +σε2

SwNw = γ2(1 — δ)2σz2 + γ2δ2N-1σ2d + N-1σε2

SxNw = γ[(1 — δ)σz2 N -1δσd2]

SyNw = γ[(β + γ)(1 — δ)σZ N-1βδσd] + N-1σε2


SxNx = σ2d + σz2


SxNz = σz2


SyNx = βσd2 +(β+γ)σz2


SN
zw


= γ(1 — δ)σz2


These results are used in the next sections to analyze the question of how
ʌ ʌ

well does the OLS estimator b (and by comparison θ) measure the true β.



More intriguing information

1. The value-added of primary schools: what is it really measuring?
2. Should Local Public Employment Services be Merged with the Local Social Benefit Administrations?
3. The name is absent
4. The name is absent
5. Incorporating global skills within UK higher education of engineers
6. Globalization, Divergence and Stagnation
7. The name is absent
8. Effort and Performance in Public-Policy Contests
9. Modeling industrial location decisions in U.S. counties
10. Importing Feminist Criticism