How to get the same point estimate using ivreg2 and xtabond2 with a constant term

[Alex Weng]

Dear all,

I am writing a project using IV as my main strategy, and I want to use a dynamic panel as a robustness check.

If we check these two following examples, they have the same results, a similar example from the xtabond2 file. But since my main IV strategy used constant term, is there a way to use the constant term in xtabond2 still?

If I get rid of "nolevel" option, the constant will show up, but it's not the same estimation as IV anymore, hence the results are also different.

use http://www.stata-press.com/data/r7/abdata.dta

ivreg2 D.n (DL.n = l2.n ) if inlist(year,1978), nocons first saverf

xtabond2 n L.n if inlist(year,1978) , gmm(L.n ) h(1) nolevel small svmat svvar


what I want to know is if there is a way to have the same results as IV regression using xtabond2 with a constant term. I.e.,
ivreg2 D.n (DL.n = l2.n ) if inlist(year,1978), first saverf


Thank you so much for your help,
Alex

[Sebastian Kripfganz]

The two commands in your example only give the same results because you restricted the estimation sample to a single year, which is no longer a panel regression.

You can do the following using my xtdpdgmm command:

webuse abdata
ivreg2 D.n (DL.n = L2.n)
xtdpdgmm D.n DL.n, iv(L2.n)

Note that the constant term in this regression refers to the first-differenced model, which would correspond to a linear time trend in the level model. If you want a constant for the level model instead, you can the following:

ivreg2 D.n (DL.n = L2.n), noconstant
xtdpdgmm n L.n, model(diff) iv(L2.n)