[Gretl-users] Re: Alternative for First Difference GMM
Am 08.04.2021 um 18:12 schrieb 3J LEMA:
My panel data have these:
Number of Companies = 10,
Time per Company =9,
Number of Independent Variables = 6
I used First Difference GMM with results as follows:
*1. Number of instruments =71 while the number of cross-sectional units
= 9.*
2. The coefficient of the lag of the DV which is used as an independent
variable is less than 1 and statistically significant (p<.001).
3. AR(1) is significant (p<.05)
4. AR(2) is nonsignificant (p>.05)
*5. Sargan Test is significant (p<.05)*
6. Wald Test is significant (p<.01).
Since #1 and #5 are not OK, then First Difference GMM is inappropriate.
I tried the Two-Step Difference and the Systems GMM and all are not OK.
Can you suggest an alternative?
Well, yes and no. In general, and not gretl-specific, the choice of the
instrument set is notoriously difficult if you don't have a clear
guidance from the theory of the subject matter. In that sense we don't
have a magic solution for you.
In terms of how to manipulate the instrument set in a dynamic panel GMM
context, the fine tuning is not implemented in the graphical interface
(AFAIK), so you would have to do a little bit of scripting to dig
deeper. Please see the reference of the dpanel command, and ch. 24 of
the user guide which is dedicated to dynamic panels.
cheers
sven