Good evening,

I am trying to estimate a dynamic panel data model, with n=6 and t=25. Hausman test determined I should carry out a fixed-effects regression. After detecting heteroscedasticity and autocorrelation, I carried out the regression with xtpcse. However, as you all might know, the results for a dynamic model with fixed effects might not be robust due to Nickell bias. My supervisor recommended the Anderson-Hsiao method, which I've tried implementing with xtivreg2, fd, instrumenting for my lagged dependent variable with different numbers of its own lags. However, after trying instrumenting with lag2 & lag3, lag3&lag4, lag2 & lag3 & lag4 and even up to l5, my regression fails to pass the weak identification test (the (Cragg-Donald Wald F statistic is always smaller or slightly larger than the 25% maximal IV size critical value of the Stock-Yogo weak ID test), so I've sort of given up on that.

My supervisor also said that I could use the "simple first difference estimator", which is, as far as I understand, just an OLS regression of the first-difference dependent variable on the first-differenced independent variables. How can I get this estimator? Would it be possible by #xtreg d.y d.l.y d.x, fe vce(cluster)# ?

Thank you!