I understand that typically the covariance matrix should not be singular (see e.g. this discussion here: Could the covariance matrix of the moment conditions in GMM be ill-conditioned?)
But in the documentation of Mike Cliff's GMM package for Matlab (https://sites.google.com/site/mcliffweb/programs; https://sites.google.com/site/mcliffweb/filebox/gmmdoc.pdf?attredirects=0) there is the following paragraph (page 8):
This is quite confusing to me. No qualification w.r.t.weighting matrix optimality is in aforementioned question on Stackoverflow. So which of these two statements is correct, or what am I missing here?
