Thanks for your reply, Christer. I guess I was not exact enough.
What I have are price and characteristics observations for many
different products, similar to the example-data of 1978 automobiles,
plus information to which of the five segments ("groups") of the
market they belong. I want to perfom a hedonic regression for every
segment, with restrictions across equations on some, but not all of
the coefficients. I thought therefore I needed to estimate some
system of related equations instead of five completely unrelated
regressions...
SUR is going to assume that you have T obs on each of the five dependent
variables. My guess is that you have the data in 'long' format, with all
five groups stacked up vertically. When you break them out into five
separate dep vars they are disjoint sets--each obs is in one of the five,
so none are in the same range of obs 1/T. Unless by chance there are
roughly the same number of obs in each of the five categories, SUR will not
work, since it assumes that the resid cov mtx can be calculated from 5
vectors of T obs.
