I mentioned in an earlier posting that a new version of ivreg2 is now
available. Mark Schaffer determined Friday that _robust, a routine in
official Stata which we employ in ivreg2, was misbehaving when time series
operators are involved. That causes ivreg2,gmm in that context to be
"thoroughly screwed up" (in technical terms) and ivreg2,robust to be
"slightly screwed up". This only affects ivreg2 when time series operators
are used. If the actual lagged or differenced variables are generated and
listed, the bug will not bite. We expect that StataCorp will have a fix (in
the form of an official update) for this very soon, as _robust may be used
in a number of contexts. In the meanwhile, please note that erroneous
results from ivreg2 in this special case are outside the authors' control.
