5:56 a.m.
Folks,
after releasing 1.7.0 a few problems have emerged. Here's my proposal on
how to deal with them:
1) There's a horrendous bug in garch that leads to bogus results if p!=q;
it affects 1.7.0, while earlier versions are ok. It's my fault that this
had gone unnoticed so far (thanks to Lee Adkins for spotting the bug). A
fix for this is already in CVS.
2) Diagnostics for models estimated via tsls leave a lot to be desired. A
version of Godfrey's (1994) autocorrelation test is already in my own
private tree; I'm waiting until Allin's back for his review. However, I'd
also like to implement Pesaran & Taylor (OBES, 1999) heteroskedasticity
test, which shouldn't take long.
3) Not urgent: I've written 2 functions for matrix import/export. Those
are sitting in my private tree right now and I'll submit the to Allin's
attention when he's back.
4) There are a few bugs in the sourceforge tracking system that need to be
addressed; none of them is lethal, but still it'd be nice to have them
fixed asap.
Since 1) is a potentially disruptive bug, I'd like to propose a bugfix
release as soon as possible, fixing 4) as far as possible.
Opinions?
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
6:28 a.m.
On Fri, Dec 07, 2007 at 11:56:21AM +0100, Riccardo (Jack) Lucchetti wrote:
Since 1) is a potentially disruptive bug, I'd like to propose a bugfix
release as soon as possible, fixing 4) as far as possible.
Opinions?
If the bugfix release is taking more than a few days to go out, maybe it's worth
to post a warning message about 1) on the gretl-announce ML (provided you have
the power to do it).
On the long term, it would be nice to have some kind of regression[1] test suite
to identify such problems before releasing; don't know if the NIST test will do,
maybe we need a carefully crafted gretl script.
Cri
[1] Of course I mean both software engineering regressions and econometric regressions :)
--
GPG/PGP Key-Id 0x943A5F0E - http://www.linux.it/~cri/cri.asc
Free software, free society - http://www.fsfeurope.org
5:51 a.m.
On Fri, 7 Dec 2007, Riccardo (Jack) Lucchetti wrote:
after releasing 1.7.0 a few problems have emerged. Here's my
proposal on how to deal with them:
1) There's a horrendous bug in garch that leads to bogus results
if p!=q; it affects 1.7.0, while earlier versions are ok. It's
my fault that this had gone unnoticed so far (thanks to Lee
Adkins for spotting the bug). A fix for this is already in CVS.
2) Diagnostics for models estimated via tsls leave a lot to be
desired. A version of Godfrey's (1994) autocorrelation test is
already in my own private tree; I'm waiting until Allin's back
for his review. However, I'd also like to implement Pesaran &
Taylor (OBES, 1999) heteroskedasticity test, which shouldn't
take long.
3) Not urgent: I've written 2 functions for matrix
import/export. Those are sitting in my private tree right now
and I'll submit the to Allin's attention when he's back.
4) There are a few bugs in the sourceforge tracking system that
need to be addressed; none of them is lethal, but still it'd be
nice to have them fixed asap.
Since 1) is a potentially disruptive bug, I'd like to propose a
bugfix release as soon as possible, fixing 4) as far as
possible.
Yes, I'd like to put out a 1.7.1 release as soon as we can.
Realistically, that would be at the end of next week (say, Friday
December 14).
In response to Cri: We do actually have a rather large
"regression suite" that I run before releases. However, it's not
quite comprehensive, and GARCH with anything other than p = q = 1
has not been on the list up till now. I'll make sure it gets
added.
Allin.
6189
days inactive
6190
days old
2 comments
3 participants
participants (3)
-
Allin Cottrell -
Cristian Rigamonti -
Riccardo (Jack) Lucchetti