Hello everyone, I'm conducting a dynamic panel data analysis using the Arellano-Bond estimator (xtabond) in Stata. My dataset includes 30 countries over a 13-year period. I want to apply both one-step and two-step GMM estimations.
However, I encounter the following problem:
The number of instruments generated by the model exceeds the number of groups.
As you know, this violates the rule of thumb that the number of instruments should be less than or equal to the number of groups (N = 30 in my case). Since I am using xtabond (not xtabond2), I cannot use options like collapse or laglimits() to reduce the number of instruments. I also cannot specify GMM or IV instruments manually in xtabond, as it's more limited than xtabond2.
Is there any recommended way to reduce the number of instruments while using xtabond? Or is it necessary to switch to xtabond2 to manage instrument proliferation?
Any suggestions or insights would be greatly appreciated.
To deal with too many instruments in xtabond GMM estimation, researchers can limit the number of lags used in instrument creation, collapse instruments using the collapse option in Stata, or reduce the number of moment conditions, as too many instruments can overfit endogenous variables and weaken the Hansen test's reliability.
To manage instrument proliferation in xtabond GMM estimation, one should reduce the number of lagged instruments, apply the collapse option to compact the instrument matrix, limit the time periods used, or use principal component analysis, all of which help prevent overfitting and improve test validity.
Chuck A Arize Thank you very much for your detailed response. I see that most of the suggested solutions such as using collapse, limiting lags with laglimits(), or controlling moment conditions, are applicable in xtabond2, not in xtabond, which I am currently using. Since xtabond does not support options like collapse, I’m wondering if there is any workaround within xtabond itself, or if switching to xtabond2 is the only viable option to control the number of instruments.
Thanks again for your help.
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity