Post Estimation using xtlsdvc estimator

[Trinil Arimurti]

Hi,
I want to estimate dynamic panel model with large T (55) and small N (17) (I have balanced panel data) using xtlsdvc code in Stata. However,

the post estimation from Bruno (2005)

is not so clear to me, in particular

on how to

test for the autocorrelation and overidentifying restriction. Is there a way to do these tests after the estimation? Thank you very much.

[Dario Maimone Ansaldo Patti]

My understanding is that xtlsdvc helps you in correcting the bias (the Nickell bias) due to the presence of the lagged dependent variable among the regressors. It achieves such a result by using a gmm estimation to define the initial value to generate such a correction (see the explanation about initial in the command help). Therefore, you have a FE estimation corrected using the procedure suggested by Kiviet (1995). You do not have a gmm estimation. Hence you do not need to test for autocorrelation and/or for overidentifying restrictions. Another point is whether you want to look at the estimation used to initialize the value to correct for the bias. In that case you may want to look at the ado file to see which estimation is invoked when the initial option is specified (specifically how many instruments are adopted) and you may want to replicate such an estimation using the xtabond2 command (which by the way is the routine used within xtlsdvc).

[Trinil Arimurti]

Thank you so much for your short explanation.

It did provide the estimation results from the xtabond estimator when we ask to display it on the output, together with the overidentifying restriction and autocorrelation test results, which is a bit confusing for me.

I will do more readings to understand more about the detail of this xtlsdvc procedure. Thank you, I really appreciate your suggestions.