Re: [Gretl-users] Simulated annealing
On Wed, 5 Oct 2011, Giuseppe Vittucci wrote:
to estimate PSTR models:
http://ideas.repec.org/p/uts/rpaper/165.html
a concentrated SSR must be minimized and the authors suggest to choose
the starting values by means of simulated annealing.
Reading the manual, I discovered that gretl is already able to use
simulated annealing to adjust the initialization in case of ML estimates
in the restrict command via the option --jitter (btw: such option is not
reported in the description of the command in the command reference).
Is there a way to use such method in gretl for initializing a generic
function to be minimized/maximized?
At present, there isn't. But I guess it wouldn't be too difficult to
generalize what we currently have in place for VECM restrictions.
This is something we could look into after gretl 1.9.6 is released
-- right now our efforts are directed towards that.
BTW, it's true, the --jitter option is not mentioned in the command
help for "restrict", but we do add the sentences, "Note that some
special considerations apply in the case of testing restrictions on
Vector Error Correction Models. Please see the Gretl User's Guide
for details." Since --jitter is a VECM-only option, I think it's
reasonable, or at least defensible (!), that it's not documented in
the generic help.
Allin Cottrell