Spatial Panel analysis for SYS-GMM

[Gino Lotito]

Hello, I have a problem. I have my model SYS_GMM (level e diff) done by using xtabond2 of Arellano and Bond. If I want to do a Spatial Panel analysis there is not the possibility to do with SYS_GMM. If i use spregdpd command i just can run(xtabond), same for spregxt... I don't know how to do?
My GMM model is like this:

xtabond2 dependent l.dependet explanatories control tau2003-tau2011, ///
gmm(dependent explanatories, lag(2 2) eq(diff)) gmm(dependent explanatories, lag(1 1) eq(lev)) ///
iv (controls tau2003-tau2011, eq(both)) h(2) ar(2) two robust noconst

I have 103 Italian Province and 10 years (2003-2012)... with GMM I'm using 2003-2011
Many thanks

[Sebastian Kripfganz]

If you generate your spatially lagged variables as new variables first, then you can use xtabond2 (or xtdpdgmm for that matter) and simply specify the instruments as needed.

[Gino Lotito]

Thanks, Sebastian, sorry for the ignorance but, how I can generate my lagged variables spatially? It's ok by doing like this?

splagvar dependent, wname(mymatrice) wfrom(Stata) ind(explanatories) order(2)


and after, i can do again the model?
Thanks

[Sebastian Kripfganz]

I do not know the splagvar command. In Stata 15 or 16, you could use the spgenerate command. With panel data it can be a bit cumbersome because you have to generate a spatial lag for every time period and then combine all of them into one variable, e.g.

copy http://www.stata-press.com/data/r15/homicide_1960_1990.dta homicide_1960_1990.dta
copy http://www.stata-press.com/data/r15/homicide_1960_1990_shp.dta homicide_1960_1990_shp.dta
use homicide_1960_1990
xtset _ID year, delta(10)
spset
spmatrix create contiguity W if year == 1990
spgenerate Whrate = W*hrate if year == 1990
forv year = 1980(-10)1960 {
    spgenerate Whrate_temp = W*hrate if year == `year'
    replace Whrate = Whrate_temp if year == `year'
    drop Whrate_temp
}

[iris lee]

I do not know the splagvar command. In Stata 15 or 16, you could use the spgenerate command. With panel data it can be a bit cumbersome because you have to generate a spatial lag for every time period and then combine all of them into one variable, e.g.

copy http://www.stata-press.com/data/r15/homicide_1960_1990.dta homicide_1960_1990.dta
copy http://www.stata-press.com/data/r15/homicide_1960_1990_shp.dta homicide_1960_1990_shp.dta
use homicide_1960_1990
xtset _ID year, delta(10)
spset
spmatrix create contiguity W if year == 1990
spgenerate Whrate = W*hrate if year == 1990
forv year = 1980(-10)1960 {
spgenerate Whrate_temp = W*hrate if year == `year'
replace Whrate = Whrate_temp if year == `year'
drop Whrate_temp
}

Hello Sebastian,

I'm trying to generate the spatially lagged variables and I'd like to know if there need any preset before these code such as shp2dta to convert shape boundary files to Stata datasets.

It's first time for me do this and I met so much troubles when I use these code with my data. I'm really confused. I'll appreciate a lot if you could give me any advice.

Kind regards,
Iris

[Maxence Morlet]

Dear all,

Apologies for reviving this thread, but I have a related question:

Regarding the use of GMM in spatial models; is there any literature implying that the Nickell (1986) bias holds true for spatial lags of the dependent variable, even though spatial models are generally estimated via maximum likelihood?

Apologies if there are seminal studies discussing this topic, I am not very familiar with this literature.