Hello I have a panel data with firm-year data. number of firms approx 1000 and number of years 10

I am estimating the following:

firm_risk_at_t = firm_risk_at_t-1 + firm_stakeholder_practices_at_t + firm_stakeholder_practices_at_t-1 + stuff + error as

my coefficient of interest is really those on stakeholder practices, i.e. I can potentially justify running a model without the AR(1) term. Nevertheless, stakeholder practices is likely endogenous. As such, i am using xtabond2

I tried pooled OLS and fixed effects to estimate the above, and the difference GMM model estimate is closer to the OLS. I therefore conclude that system GMM is not really worth it and decide to use difference GMM. this is my code:

xtabond2 firm_risk L.firm_risk L(0/1).stakeholder_practices yr* , gmm(L.firm_risk) iv(stakeholder_practices yr*) robust small noleveleq orthogonal


when i run the above, i get a negative estimate on stakeholder practices.

When i run the above without the orthogonal option, i get a positive estimate.

Is there something blatantly different with that option?

Also, my p values for Sargan and Hansen tests are close to 0.000. Any suggestions?