Dear Stata users,

I have always assumed that in the presence of serial autocorrelation and assumptions of endogeneity, a correctly specified GMM model should lead to "poorer" results (in the sense of less significance) than standard xtreg/xtregar because the autocorrelation and/or endogeneity will lead to false significance in the latter (overinflated coefficients, smaller errors). Yet every now and then I see someone report poor xtreg/xtregar results but then five-star GMM results. This seems weird to me. Am I alone? Because GMM is so sensitive to assumptions and instruments and lags etc., my knee-jerk response is to think that in such cases the GMM must be flawed.

So my basic question is: could you imagine xtreg output in which not much is going on in terms of significance but then great GMM results based on the same data, and that the latter results would be "true" and not spurious? Thanks in advance for any replies.