xtabond2 - AR(1) and AR(2) tests

[Felipe Cruz]

I'm estimating a "Earnings Management" model for my thesis. I'm trying to obtain the residuals from this model in order to identify what we call as "discretionary accruasl" in accounting.
Anyway, here is my question: In the best model I can obtain via the Systemic GMM, AR(1) is negative, but not significant and AR(2) is also not significant. I know that ideally AR(1) should be negative and significant, but I guess my data is just not perfect. Does this completely invalidate my model?

[Sebastian Kripfganz]

I suppose that this is fine if everything else checks out. Since you want to work with the residuals, you could just look at some summary statistics for them or plot them to see if anything looks odd.

I would suggest to rescale your reverseasset variable (multiply it by 1 million), if meaningful, so that all of your coefficients are on a similar scale. Otherwise, this could cause computational problems.

[Felipe Cruz]

Thank you very much for your answer! I have used the residuals as the dependent variable on a different model and my findings are really satisfactory. They meet theoretical expectations quite well. I think they really do represent the variable I'm trying to capture... I'll just try to defend this model so I do not have to change the other ones. I'll also check whether changing the assets variable scale has any meaningful effect on my results. I did, however, use as many decimal cases as I could. The coefficient scale is not really an issue tho.