Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: xtlsdvc, sargan test and heteroscedastisity

From   Magnus Gulbrandsen <>
Subject   st: xtlsdvc, sargan test and heteroscedastisity
Date   Tue, 20 Dec 2011 08:14:45 +0100


I am running a FE regression with LDV, and thus need to correct for
the possible Hurwicz bias. I use the -xtlsdvc- command, with option on
initial regression -initial(ab)- . After the regressions I have a few

1) The Sargan test comes out:

                     Sargan test of over-identifying restrictions:
                    chi2(54) =   185.87      Prob > chi2 = 0.0000

and if I am correct, this means in plain words that the instruments in
the Arrelano-Bond estimation are not valid(correct?). If so, does this
mean that the estimates in the LSDVC regressions also are wrong? Are
there any solutions to this problem?

2) Are there any good guidlines for how many bootstrap repetitions one
should use in the -vcov(#)- option? I have read elsewhere (Stata
FAQ:"How large should the bootstrapped samples be relative to the
total number of cases in the dataset?") that it doesn't depend on the
number of observations, and that optimally it should be infinitely
many.  I have about 1200 observations (132 groups and 12 time
observations), and it seems to take a lot of time just doing 100

3) Econometricians seem to agree that as time observations (T)
increases the Hurwicz bias becomes negligible, but identify different
number of T where this is the case (ranging from 10 to 20). Thus, my
dataset (with T=12) is a border case. If the z-values and the
different coefficients are only marginally different in the pure
FE+LDV approach and the LSDVC approach with bootstrapped se, should I
take this as a sign that the Hurwicz bias is so small that I just as
well can carry on with the FE+LDV model?

4) I would like to test for between-group and within-group
heteroskedasticity after -xtreg, fe- .  -xttest3- to my knowledge does
the first mentioned. It indicates group-wise heteroscedasticity, even
when using robust/cluster standard errors (-vce(robust)- or
-vce(cluster clusterid)-). Any suggestions to how this can be
corrected? And, how can i check for within-group heteroscdasticity?

All responense will be greatly appreciated.

Best regards,

Magnus Gulbrandsen

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index