Re: [Gretl-users] Data error with quantreg
On Wed, 4 Jul 2012, Ashley Dunstan wrote:
Sorry can this message please be removed from the list, I meant to
send it just to Allin.
I don't have a means of removing a message from the archive, but
your dataset was not posted.
Anyway, the problem appears to be near-singularity in the design
matrix. Quantile regression can be numerically tricky at the best of
times and you're pushing it a bit, trying to estimate over 200
parameters. The Frisch-Newton algorithm is breaking down at a point
where it attempts to solve a linear system via Cholesky
decomposition; the relevant matrix is not positive definite so far
as the computer is concerned. I've modified the source in CVS so
that you get a more informative error message in this case.
I tried your problem using the latest version of Roger Koenker's
quantreg package (4.81) for R, and got the same breakdown.
Allin Cottrell