I use xtabond two-step, but I got 'Warning: Arellano and Bond
recommend using one-step results for inference on coefficients'. Could
anyone help to explain why it comes like this and how I can solve the
problems please?
It is well known that the regular GMM standard errors in two-step
estimation are biased, so that making inferences from those SEs may
be unreliable. Thus Windmeijer developed the "Windmeijer correction"
which reduces their bias. This is available with official Stata's -
xtabond- if you specify the twostep and vce(robust) options. It is
also available in the user-written -xtabond2-, where it is the
default if twostep estimation is performed. -findit xtabond2- for
details.
