[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: RE: Fixed effects decomposition (was xthtaylor)
There is a discussion of the Hausman Taylor method in Wooldridge's book, "Econometric Analysis of Cross Section and Panel Data" in pps 325-328, and the estimation is coded in stata 8.
From a fe regression, one can not obtain a(i) as they are cleared off through time demeaning ( and so are z(i)). You can estimate a(i) by using individual dummies and running an OLS, but may loose several degrees of freedom.
The advantage of H-T estimation is that you can obtain consistent and efficient estimates of g. This is done in two basic steps. A fe regression is run to obtain within group errors. From these one obtains consistent, though inefficient, parameter estimates.
From this intermediate step, one can obtain standard errors of the individual and idiosyncratic error components and run a re model to obtain consistent and efficient parameter estimates.
In comparison to this approach, I do not think that the two- step procedure that you have outlined may work.
Hope this helps,
From: email@example.com on behalf of ACavallo@lexecon.com
Sent: Tue 10-Feb-04 2:35 PM
Subject: st: Fixed effects decomposition (was xthtaylor)
Thanks for your tips - I forgot that these characteristics can be estimated
using a random effects model.
My question about the decomposition still stands though, how do I estimate
g in the model below when using fixed effects.
y(it)=b*x(it) + g*z(i) + a(i) + e(it)
where x(it) are time varying characteristics, z(i) are time invariant
characteristics, and a(i) are the fixed effects.
My proposal is to estimate fixed effects model, predict a(i), then regress
a(i) on z. I would bootstrap to get the standard errors right on the
xtreg y x, fe
predict a, u
by id : keep if _n==1
regress a z
I looked in Greene's "Econometric Analysis" and Baltagi's "Econometric
Analysis of Panel Data" but did not see any explanation. Does anyone have
references or comments on my proposal?
(312) 322-0208 voice
(312) 322-0218 fax
>Can I use XTHTAYLOR assuming no variables are correlated with a(i)? In
>other words, is the endog(varlist_endog) option required? I don't yet
>Stata 8 so I can't just try this.
The online help for -xthtaylor- states that the endog() option is required,
However, the Hausman-Taylor estimator of a model where every variable in X
and Z is assumed to be uncorrelated with the random effect [a(i), or u(i)
in stata's notation] is simply a random-effects model (-xtreg, re- y on X
and Z varlists). I don't have the paper here, but I think this is stated in
Hausman and Taylor (1981) econometrica paper.
>If not, does anyone have a reaction to this proposed method:
> 1. estimate y(it)=b*x(it) + a(i) + e(it)
> 2. regress ahat(i) on z(i) to estimate ghat, using bootstrap to get
>standard errors right
Don't trust me much in this, but I think that depending on the procedures
used in steps (1) and (2) (and possibly on the (un)balanced nature of your
panel data set) you could get a "ghat" estimator that would be consistent
but not efficient. If all variables in X and Z are really exogenous,
-xtreg, fe- is both consistent and efficient.
Hope this helps,
Mario F. Rueda Narváez
Departamento de Estadística y Econometría
Facultad de Ciencias Económicas
Universidad de Málaga
El Ejido s/n
29013 Málaga (España)
Tlf: +34 952 13 71 90
Fax: +34 952 13 72 62
* For searches and help try: