12:37 p.m.
Hi all,
a strange thing just happened to me while looking at some data in gretl:
After hitting some key combinations (wrong ones), gretl suddenly started
reading the numbers -- I mean, literally reading, with an American
accent, in English, over the loudspeaker.
And closing gretl didn't stop it! I had to kill the gretlw32.exe process
(on Windows, obviously) in the task manager...
That was really spooky...
What's behind this?
-Sven
12:48 p.m.
On Tue, 15 Aug 2006, Sven Schreiber wrote:
a strange thing just happened to me while looking at some data in
gretl:
After hitting some key combinations (wrong ones), gretl suddenly started
reading the numbers -- I mean, literally reading, with an American
accent, in English, over the loudspeaker.
And closing gretl didn't stop it! I had to kill the gretlw32.exe process
(on Windows, obviously) in the task manager...
The key for starting gretl reading is 'a' for audio. To stop
it, press 'x' over the window in question.
Allin Cottrell
8:54 a.m.
New subject: VECM estimation
Hi everybody,
gotta one request about VECM estimation. Once, you said that you will
finished VECM estimation in gretl i.e. that it will be possible to impose,
not only to test, restrictions on alpha and beta space.
Also, Jack said that it will include the possibility to check Granger
causality in VECM.
Maybe I'm a little bit pushy but when can this be expected?
Thanx
_________________________________________________________________
Be the first to hear what's new at MSN - sign up to our free newsletters!
http://www.msn.co.uk/newsletters
12:25 p.m.
New subject: VECM estimation
On Thu, 17 Aug 2006, john w wrote:
gotta one request about VECM estimation. Once, you said that you will
finished VECM estimation in gretl i.e. that it will be possible to impose,
not only to test, restrictions on alpha and beta space.
Also, Jack said that it will include the possibility to check Granger
causality in VECM.
Maybe I'm a little bit pushy but when can this be expected?
I would guess, by the end of this year.
At present we have the ability to test homogeneous restrictions
on beta, the cointegrating vector(s), only -- either using the
GUI or the "restrict" command following a vecm. Thanks to a
pointer from Sven I now have a reasonable idea of how to proceed
with adding more general tests on both beta and alpha (the
"adjustments" vector), at least in principle. But it's a fairly
substantial task.
Allin.
12:40 p.m.
New subject: VECM estimation
Allin Cottrell schrieb:
On Thu, 17 Aug 2006, john w wrote:
At present we have the ability to test homogeneous restrictions on beta,
the cointegrating vector(s), only -- either using the GUI or the
"restrict" command following a vecm. Thanks to a pointer from Sven I
now have a reasonable idea of how to proceed with adding more general
tests on both beta and alpha (the "adjustments" vector), at least in
principle. But it's a fairly substantial task.
Yes I also think it would be a fair amount of work.
What is not so difficult is to test homogeneous restrictions on alpha,
e.g. for weak exogeneity tests. Once I'm familiar with the new matrix
programming facilities in gretl, I might be able to contribute that
function.
I'm also interested in replicating the functionality of pcgive in terms
of general cointegration restrictions, and maybe I could help with
prototyping (I use numpy -- numerical python). But I don't have time
before October.
Cheers,
Sven
1:03 p.m.
New subject: VECM estimation
Thnx Sven.
From: Sven Schreiber <svetosch(a)gmx.net>
Reply-To: Gretl list <gretl-users(a)ricardo.ecn.wfu.edu>
To: Gretl list <gretl-users(a)ricardo.ecn.wfu.edu>
Subject: Re: [Gretl-users] VECM estimation
Date: Mon, 21 Aug 2006 14:40:43 +0200
Allin Cottrell schrieb:
> On Thu, 17 Aug 2006, john w wrote:
>
> At present we have the ability to test homogeneous restrictions on beta,
> the cointegrating vector(s), only -- either using the GUI or the
> "restrict" command following a vecm. Thanks to a pointer from Sven I
> now have a reasonable idea of how to proceed with adding more general
> tests on both beta and alpha (the "adjustments" vector), at least in
> principle. But it's a fairly substantial task.
>
Yes I also think it would be a fair amount of work.
What is not so difficult is to test homogeneous restrictions on alpha,
e.g. for weak exogeneity tests. Once I'm familiar with the new matrix
programming facilities in gretl, I might be able to contribute that
function.
I'm also interested in replicating the functionality of pcgive in terms
of general cointegration restrictions, and maybe I could help with
prototyping (I use numpy -- numerical python). But I don't have time
before October.
Cheers,
Sven
_______________________________________________
Gretl-users mailing list
Gretl-users(a)ricardo.ecn.wfu.edu
http://ricardo.ecn.wfu.edu/mailman/listinfo/gretl-users
_________________________________________________________________
Be the first to hear what's new at MSN - sign up to our free newsletters!
http://www.msn.co.uk/newsletters
1:12 p.m.
New subject: VECM estimation
Sven Schreiber schrieb:
Allin Cottrell schrieb:
> On Thu, 17 Aug 2006, john w wrote:
> At present we have the ability to test homogeneous restrictions on beta,
> the cointegrating vector(s), only -- either using the GUI or the
> "restrict" command following a vecm. Thanks to a pointer from Sven I
> now have a reasonable idea of how to proceed with adding more general
> tests on both beta and alpha (the "adjustments" vector), at least in
> principle. But it's a fairly substantial task.
>
Yes I also think it would be a fair amount of work.
What is not so difficult is to test homogeneous restrictions on alpha,
e.g. for weak exogeneity tests. Once I'm familiar with the new matrix
programming facilities in gretl, I might be able to contribute that
function.
I'm also interested in replicating the functionality of pcgive in terms
of general cointegration restrictions, and maybe I could help with
prototyping (I use numpy -- numerical python). But I don't have time
before October.
On second thought, I think general linear restrictions (as in 4.4 of the
paper) are normally enough, although I occasionally needed nonlinear
restrictions in the past as well. But the linear stuff is also what
Eviews does, and is probably a more realistic goal in the medium term.
I guess the nice matrix and scripting capabilities of gretl would also
be enough to make that work, so I could help with the prototype. It
would be useful, however, if one could access some preliminary
variables/matrices from the Johansen procedure (like S00, S11 etc.) from
a script, similar to what $jalpha apparently does now. I mean not
officially and not documented, but just so I can help without starting
from scratch. Would that be easy to do for you Allin, or would it just
shift workload from me to you?
And how actually does it work now to test the homogeneous restrictions
you mentioned, e.g. via GUI? I wasn't obvious to me.
Thanks,
Sven
3:30 p.m.
New subject: VECM estimation
I guess the nice matrix and scripting capabilities of gretl would
also
be enough to make that work, so I could help with the prototype. It
would be useful, however, if one could access some preliminary
variables/matrices from the Johansen procedure (like S00, S11 etc.) from
a script, similar to what $jalpha apparently does now. I mean not
officially and not documented, but just so I can help without starting
from scratch. Would that be easy to do for you Allin, or would it just
shift workload from me to you?
And how actually does it work now to test the homogeneous restrictions
you mentioned, e.g. via GUI? I wasn't obvious to me.
the matrices S00 & co. are quite easy to build by combining lists, loop, genr
and matrix functions; all in all, you need moment matrices of residuals. On
the other hand, I've modified the code for vecm estimation (patch attached,
Allin) so to enable the user to retrieve the covariance matrix of the
estimated beta (after triangular normalisation), which is NOT trivial to
compute. This makes it possible to compute Wald tests on cointegration
vectors, JMulTi style (but see below).
With the attached patch, here comes an example, which uses (surprise,
surprise) Johansen's Denmark example data: we test two restrictions: unit
elasticity wrt income and that the spread between the two rates measures the
cost of holding money.
Note that we can already do this via a LR test. A Wald test, however, can be
used for nonlinear restrictions too, via the delta method.
# ------------------- begin example script -----------------------------
open denmark
vecm 2 1 1 2 3 4 --rc --seasonal
matrix b = $jbeta
matrix V = $jvbeta
matrix R = { 1, 1, 0, 0, 0 ; 0, 0, 1, 1, 0 }
scalar df = rows(R)
matrix W = R * b
matrix R = R[,2:5]
matrix V = R * V * R'
scalar WT = W' * inv(V) * W
printf "Wald test: %g (%d df, pvalue = %g)\n", WT, df, pvalue(X,WT,df)
# -------------------- end example script ------------------------------
While I tested this, I realised that I couldn't reproduce Johansen's results
with JMulTi; either I'm missing something, or JMulTi's VECM estimation goes
through a different algorithm from ours. We do VECMs as per Johansen's book;
so does PcGive, and I believe there should be no differences between our
results and what PcGive yields. But then again, Microfit seems to differ too
(I don't have Microfit, so I can't run an in-depth comparison).
Allin, if you approve the patch, I'll update the manual.
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona
7:31 p.m.
New subject: VECM estimation
Riccardo Jack Lucchetti schrieb:
the matrices S00 & co. are quite easy to build by combining lists, loop, genr
and matrix functions; all in all, you need moment matrices of residuals. On
Sure, but dealing with sample lengths, lagged differences, exogenous
variables etc. is not fun and it's already implemented in a nice dialog,
so why not use it?
Note that we can already do this via a LR test.
Where is that documented?
While I tested this, I realised that I couldn't reproduce
Johansen's results
with JMulTi; either I'm missing something, or JMulTi's VECM estimation goes
through a different algorithm from ours. We do VECMs as per Johansen's book;
so does PcGive, and I believe there should be no differences between our
results and what PcGive yields.
I have all those programs, do you want me to test/confirm something?
cheers,
sven
2:50 a.m.
New subject: VECM estimation
On Wed, 30 Aug 2006, Riccardo Jack Lucchetti wrote:
I've modified the code for vecm estimation (patch attached,
Allin) so to enable the user to retrieve the covariance matrix
of the estimated beta (after triangular normalisation), which
is NOT trivial to compute. This makes it possible to compute
Wald tests on cointegration vectors, JMulTi style (but see
below).
Nice, I'm applying the patch. I'll take a look at the comparison
issue with JMulTi before long.
Allin.
7:23 p.m.
New subject: VECM estimation
While I tested this, I realised that I couldn't reproduce Johansen's results
with JMulTi; either I'm missing something, or JMulTi's VECM
estimation goes
through a different algorithm from ours. We do VECMs as per Johansen's
book;
so does PcGive, and I believe there should be no differences between our
results and what PcGive yields.
I already tested. Results i.e. estimation is the same in JMulTi and gretl.
You probabbly missed something when imposing cases in JMulTi. For example
you maybe entered lags in JMulTi's VECM in levels and you should in
differences. This could be one of the reasons.
But the results are exactly the same!
Also, in JMulTi you can impose restrictions (i.e. you can't reestimate model
with imposed restrictions on beta) in Johansen estimation like in for
example PcGive or Eviews. It's explained in help files. This feature is not
implemented and it's a little bit strange because this kind of estimation is
a "benchmark" technicque.
If it will be implemented in gretl this will be a very big plus!
If you want I could post how one should specify Johansen cases in JMulTi.
_________________________________________________________________
Windows Live Messenger has arrived. Click here to download it for free!
http://imagine-msn.com/messenger/launch80/?locale=en-gb
2:40 a.m.
New subject: VECM estimation
On Wed, 30 Aug 2006, Sven Schreiber wrote:
On second thought, I think general linear restrictions (as in
4.4 of the paper) are normally enough...
I guess the nice matrix and scripting capabilities of gretl
would also be enough to make that work, so I could help with
the prototype. It would be useful, however, if one could
access some preliminary variables/matrices from the Johansen
procedure (like S00, S11 etc.) from a script, similar to what
$jalpha apparently does now.
That shouldn't be too hard to arrange, but it'll take me a day
or two as I'm pressed by other things right now.
And how actually does it work now to test the homogeneous
restrictions you mentioned, e.g. via GUI? I wasn't obvious to
me.
I need to check my own sample scripts and Hamilton's book again;
I'll do that tomorrow and post again. There should be an
example of this sort of thing in the manual (assuming it
actually works at present!).
Allin.
3:36 a.m.
New subject: VECM estimation
On Wed, 30 Aug 2006, Sven Schreiber wrote:
And how actually does it work now to test the homogeneous
restrictions you mentioned, e.g. via GUI? I wasn't obvious to
me.
Yes, it needs to be documented properly, but here's an example
from Chapter 20 of Hamilton's "Time Series Analysis". The data
file, hamilton.gdt, is supplied with the gretl distribution.
Hamilton's variables are the logs of the US price level, p, the
the US/Lira exchange rate, s, and the Italian price level, pf,
(all monthly, *100, and taken as offsets relative to 1973:01).
He runs a VECM and reports the cointegrating vector, beta, as
(1.00, -0.04, -0.56)'.
Gretl gets:
p(-1) 1.0000
(0.0000)
s(-1) -0.036957
(0.028013)
pf(-1) -0.55680
(0.014770)
OK so far. He then goes on to talk about LR tests regarding the
cointegrating vector.
Test 1: Is the middle coefficient zero? (That is, the
cointegration involves only the two price levels.)
Hamilton works this through to an LR Chi-square(1) of
0.97. I think this is a little approximate. With
gretl we do
restrict
b2 = 0
end restrict
and get
2 * (lu - lr) = 0.97682
P(Chi-Square(1) > 0.97682 = 0.322985
I think this is right.
Test 2: The real exchange rate is stationary, or beta is
proportional to (1, -1, -1)'. In gretl we do
restrict
b1 - b2 = 0
b1 - b3 = 0
end restrict
and get
2 * (lu - lr) = 13.9243
P(Chi-Square(2) > 13.9243 = 0.000947048
Hamilton reports a test statistic of 13.92, so we're on the
money.
Full script:
open hamilton.gdt
genr p = 100*(log(PZUNEW)-log(PZUNEW[1973:01]))
genr pf = 100*(log(PC6IT)-log(PC6IT[1973:01]))
genr s = -100*(log(EXRITL)-log(EXRITL[1973:01]))
# The sample period used by Hamilton
smpl 1974:2 ;
Hamilton <- vecm 12 1 p s pf
restrict Hamilton
b2 = 0
end restrict
restrict Hamilton
b1 + b2 = 0
b1 + b3 = 0
end restrict
--
Allin
1:22 a.m.
New subject: VECM estimation
On Thu, 31 Aug 2006, I wrote:
Test 2: The real exchange rate is stationary, or beta is
proportional to (1, -1, -1)'. In gretl we do
restrict
b1 - b2 = 0
b1 - b3 = 0
end restrict
Hmm, that would not be correct. There's now a correct version
of the hamilton VECM script in gretl CVS. The test above should
be:
restrict
b1 + b2 = 0
b1 + b3 = 0
end restrict
Allin.
5:24 p.m.
Allin Cottrell schrieb:
On Tue, 15 Aug 2006, Sven Schreiber wrote:
> a strange thing just happened to me while looking at some data in gretl:
> After hitting some key combinations (wrong ones), gretl suddenly started
> reading the numbers -- I mean, literally reading, with an American
> accent, in English, over the loudspeaker.
>
> And closing gretl didn't stop it! I had to kill the gretlw32.exe process
> (on Windows, obviously) in the task manager...
The key for starting gretl reading is 'a' for audio. To stop it, press
'x' over the window in question.
Thanks for the hint. If possible, however, I think that closing gretl
should also stop the reading (and other gretl-related stuff I don't know
nothing about yet).
-Sven
6693
days inactive
6715
days old
14 comments
4 participants
participants (4)
-
Allin Cottrell -
john w -
Riccardo Jack Lucchetti -
Sven Schreiber