xtabond2 with bootstrap

[Muhammad Anees]

Hello

I am using Stata 14.1 on a Windows 10 machine with i7, xtaboad2 and bootstrap, I would like your suggestions and comments to improve the following. The earlier posts on this issue also does not reveal any solution so far.

The following code is used but that does not work and give results in the warning/error as shown. I would appreciate if I can get a hint to reolve this issue and get the bootstrapping results.

. bootstrap, reps(1000): xtabond2 n l1.n k rec, gmm(l1.n) iv(k rec) twostep robust small
time-series operators are not allowed with bootstrap without panels, see tsset
r(198);

. bootstrap, reps(1000): xtabond2 n n k rec, gmm(n) iv(k rec) twostep robust small
(running xtabond2 on estimation sample)

Bootstrap replications (1000)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 50
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 100
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 150
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 200
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 250
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 300
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 350
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 400
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 450
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 500
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 550
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 600
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 650
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 700
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 750
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 800
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 850
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 900
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 950
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 1000
insufficient observations to compute bootstrap standard errors
no results will be saved
r(2000);

[Jorge Eduardo Perez Perez]

Right now, your bootstrap is resampling individual observations, not panels, breaking the time structure of the dataset. You need to use clustered bootstrapping, see the FAQ:

http://www.stata.com/support/faqs/st...th-panel-data/

[Muhammad Anees]

Dear Jorge,

I did try the FAQ and other discussions on previous Statalist archives. All these gives the similar error.

I did run the suggested FAQ as:

xtset id
xtabond2 n l1.n k rec, gmm(l1.n) iv(k rec) vce(bootstrap (_b[k] - _b[rec]),rep(10) seed(123))
program my_xtboot,rclass
summarize d.`1',meanonly
scalar mean`1'=r(mean)
summarize d.`2',meanonly
scalar mean`2'=r(mean)
return scalar ratio=scalar(mean`1')/scalar(mean`2')
end
generate newid = id
tsset id year
generate sample=1-missing(rec,ys)
keep if sample
bootstrap ratio=r(ratio),rep(10) seed(123) cluster(id) idcluster(newid) nowarn:my_xtboot rec ys

and the that also gives the error as:

Bootstrap replications (10)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
repeated time values within panel
the most likely cause for this error is misspecifying the cluster(), idcluster(), or group() option

Did I work wrong something? As I changed the cluster and idcluster with alternates and still the same warning.

I wish this can be helped.

Regards

[Jorge Eduardo Perez Perez]

I see you were trying to replicate the example in the FAQ. You need to tsset (xtset) by newid before bootstrapping, not by id:

use http://www.stata-press.com/data/r14/nlswork, clear
xtset idcode year
cap program drop my_xtboot
program my_xtboot,rclass
summarize d.`1',meanonly
scalar mean`1'=r(mean)
summarize d.`2',meanonly
scalar mean`2'=r(mean)
return scalar ratio=scalar(mean`1')/scalar(mean`2')
end
generate newid = idcode
xtset newid year
generate sample=1-missing(ttl_exp,hours)
keep if sample
bootstrap ratio=r(ratio),rep(10) seed(123) cluster(idcode) idcluster(newid) nowarn:my_xtboot ttl_exp hours

[Muhammad Anees]

Dear Jorge,

Thanks for your response. No, I did not try to replicate the the FAQ but wished to transfer the FAQ to my problem. May be I did not make it clear enough in my original post. I wish to bootstrap results from the xtabond2 which I cant do it with bootstrap prefix as you can see in the original post. The problem is there were a few past discussions and these were also not completely resolved as the results from those examples were also not possible to be replicated.

Can you help me how to bootstrap results from xtabond2?

[Jorge Eduardo Perez Perez]

I am reluctant to keep advising on this because a quick search about bootstrapping on dynamic panel data models did not yield any application of the full nonparametric bootstrap, even with clustering, for these models. I found a few papers that discuss parametric alternatives, e.g. http://www.sciencedirect.com/science...0440761500041X

Alternatively, you can try a least squares dummy variable estimator instead of GMM, if you have a long panel. See http://www.stata-journal.com/article...article=st0091

[Fredj Imene]

Dear Sir, I found the same problem. Have you found a solution to bootstrap in xtabond2?