how to improve Arellano-Bond test for AR(2) in first differences in system GMM?

#1
Hello,

I am estimating my regression using system GMM.

xtabond2 y L.y x1 x2 x3 yr2004-yr2011 z, gmm(y, lag(2 .) ) iv(x3 yr2004-yr2011, eq(lev)) iv(x1 x2 z) two ar(4) ort nodiff noconst rob

I am concerning most below:
- instrument number to make sure not to exceed group number
- Hangen/Sargan test
- Arellano-Bond test for AR(2) in first differences

When inst.number>grp.number, I limit the lag and collapsing them; when Hansen/Sargan test is not valid, I try different gmmstyle, ivstyle compositions. thus I can fix them to report good results. BUT I just can not improve Arellano-Bond test for AR(2) in first differences. In the all the cases, they show autocorrelation in the residuals.

What should I do? Please give me advice. Thank you very much in advance.
 

noetsi

No cake for spunky
#2
A common way to deal with AR issues in time series is to specify a AR2 lag as a predictor. This is for example the way that regression with autoregressive error works. I don't know if this applies to what you are doing or not. The test I use for detecting such errors is different (although it is not the test that is the issue, but how you address what it detects).