1:57 a.m.
Jack Lucchetti has pointed out that the fixes I committed earlier
today for the (new) "omit exogenous vars" test for VARs were not
complete: some errors remained.
I think I have now fixed the remaining errors in CVS, and in an
updated Windows snapshot.
One other thing in the Windows snapshot: I think I've fixed the
issue whereby if you chose a monochrome EMF file, you still got a
color representation of the graph.
The new lag selection mechanism still has known bugs, and I'm
working on those. The "bounding box" problem with EMFs copied to
the Windows clipboard has not been addressed yet.
Allin Cottrell
7:41 p.m.
New subject: Omit exogenous variables-OK
Now it seems that omit exogenous vars works ok.
In some cases when testing VAR lags and VAR models it seems that Gretl
crashes. I will make more tests and will post the this cases.
Can you post me the link where I can download CVS version for Windows? Thnx
In my previous posts I asked about Johansen cointegration tests with Gretl
(Post was: Cointegration test - help!). Can someone please help me or post
me some references regarding cointegration tests with Gretl. I already
explaind in what I am interested particulary.
In short sentences: What is the "procedure" or a "good way" to perform
a
Johansen cointegration test with Gretl?
Thnx in advance.
_________________________________________________________________
Are you using the latest version of MSN Messenger? Download MSN Messenger
7.5 today! http://messenger.msn.co.uk
2:18 a.m.
New subject: Omit exogenous variables-OK
On Tue, 7 Feb 2006, john w wrote:
Now it seems that omit exogenous vars works ok. In some cases when
testing VAR lags and VAR models it seems that Gretl crashes. I
will make more tests and will post the this cases.
Thanks, please do.
Can you post me the link where I can download CVS version for
Windows? Thnx
http://ricardo.ecn.wfu.edu/pub/gretl/gretl_install.exe
Or do you mean a version of CVS for Windows?
http://www.cvsnt.org/
http://www.wincvs.org/
In my previous posts I asked about Johansen cointegration tests
with Gretl (Post was: Cointegration test - help!). Can someone
please help me or post me some references regarding cointegration
tests with Gretl. I already explaind in what I am interested
particulary. In short sentences: What is the "procedure" or a
"good way" to perform a Johansen cointegration test with Gretl?
I'd recommend reading James Hamilton's Time Series Analysis on this
topic
http://pup.princeton.edu/titles/5386.html
There's not really anything specific to gretl that needs to be said,
other than what has been said by Jack in the chapter "Cointegration
and Vector Error Correction Models" in the gretl user's guide.
Allin Cottrell
6:22 p.m.
New subject: Cointegration with Gretl and errorbars
Allin:
Or do you mean a version of CVS for Windows?
http://www.cvsnt.org/
http://www.wincvs.org/
Thnx Allin.
Yws, I was interested in CVS for Windows.
Regarding Johansen cointegration tests:
I'd recommend reading James Hamilton's Time Series Analysis on
this topic
http://pup.princeton.edu/titles/5386.html
I'll check.
There's not really anything specific to gretl that needs to be
said, other
than what has been said by Jack in the chapter "Cointegration and Vector
Error Correction Models" in the gretl user's guide.
I already saw help session about cointegration.
I will give you one example..If you are testing for Johansen cointegration
rank with PcGive you firstly set up a VAR model which can be tested for no
autocorrelation, Arch etc. Then this tested model is then re-estimated in
Cointegrated VAR model i.e. VECM.
If you are testing for cointegration rank using JMulTi then you can test
using AIC and BIC criteria. There is no further output.
Gretl gives output when testing for cointegration and this is perfect.
My question is: to determine the optimal number of lags (which remooves
autocorrelation) what are your guidelines?
On what basis do you say that "this will be my optimal number of lags" in
cointegration test?
Other questions:
1. Errorbars are still set as defoult in VECM graphs and can't be changed?
2. What do you say about a option to add EC graph when VECM is estimated?
3. Maybe will be nice to put once again Alpha results beside Beta vector
results?
Sorry for so much questions.
Thank you
_________________________________________________________________
Are you using the latest version of MSN Messenger? Download MSN Messenger
7.5 today! http://messenger.msn.co.uk
8:30 p.m.
New subject: Cointegration with Gretl and errorbars
On Wed, February 8, 2006 19:22, john w wrote:
Gretl gives output when testing for cointegration and this is perfect.
My question is: to determine the optimal number of lags (which remooves
autocorrelation) what are your guidelines?
On what basis do you say that "this will be my optimal number of lags" in
cointegration test?
In my view, data analysis is more akin to a craft than to science, so
"recipes" are something I generally dislike.
However, a general strategy for estimating VECMs that I would personally
consider sensible could follow these steps:
1) A cointegrated VAR (ie a VECM) is nothing but a VAR with nonlinear
restrictions; therefore, an unrestricted VAR is a perfectly valid model. The
only thing you miss is using the additional information coming from the
restrictions on the rank of A(1), ie cointegration rank and long-run
equilibrium conditions. Clearly, these are often the things you're most
interestd in, but still the general VAR setup (lags, deterministic components)
can be decided on the basis of the unrestricted VAR result and a pinch of
economic reasoning.
2) Use Johansen's test for establishing the cointegrating rank. Again, don't
be a martyr of the p-value religion. If in doubt, aim for a sensible economic
interpretation whenever possible.
3) Estimate the VECM and do all the usual checks. If necessary, do some
testing on the cointegration vectors too (ie zero restrictions).
Unfortunately, this is an area where gretl is still a bit behind specilised
packages: it shouldn't be difficult do add Wald tests on the betas, but for
Johansen-like LR general tests you'll have to wait a bit more.
No doubts someone will disagree with the above, with very good reasons. I make
no pretence of "holding the truth", so I am open to criticism.
--
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona
9:15 p.m.
New subject: Cointegration with Gretl -Suggestions!
Thank you very much Jack!
Jack:
In my view, data analysis is more akin to a craft than to science, so
"recipes" are something I generally dislike.
I agree.
However, a general strategy for estimating VECMs that I would
personally
consider sensible could follow these steps:
1) A cointegrated VAR (ie a VECM) is nothing but a VAR with nonlinear
restrictions; therefore, an unrestricted VAR is a perfectly valid model.
The
only thing you miss is using the additional information coming from the
restrictions on the rank of A(1), ie cointegration rank and long-run
equilibrium conditions. Clearly, these are often the things you're most
interestd in, but still the general VAR setup (lags, deterministic
components)
can be decided on the basis of the unrestricted VAR result and a pinch of
economic reasoning.
Again, I tottaly agree with you Jack.
The main problem is to determine "well" the wright cointegration rank and
this is close related and very sensitive to number of lags!
I'm giving a suggestion. Listen carefully-as you said a VECM is a VAR VAR
with nonlinear restrictions. This is well solved in PcGive. When selecting
data which enter in VAR ( which will be the basis to determine the number of
cointegration rank for future VECM) you can choose if constant and trend
will be unrestricted or restricted i.e. exogenous. This means that you can
enter (in a VAR model) constant as unrestricted but also as exogenous. This
stays for trend too.
So, you are getting Johansen cases:
Const, Trend - all as restricted
Const - unrestricted, Trend - restricted
Const - restricted, Trend - unrestricted etc.
After such estimation in which you can determine the constant and trend to
be unrestricted or not you procede with usual VAR model and usual testings.
This results are then applyed for VECM estimation.
So, I think that it will till be a great option to have a possibility to
determine constant and trend as unrestricted or restricted.
What do you think?
This will be a perfect feature. Plus, very proffesional.
_________________________________________________________________
Be the first to hear what's new at MSN - sign up to our free newsletters!
http://www.msn.co.uk/newsletters
9:37 p.m.
New subject: Cointegration with Gretl -Suggestions!
On Wed, February 8, 2006 22:15, john w wrote:
I'm giving a suggestion. Listen carefully-as you said a VECM is a
VAR VAR
with nonlinear restrictions. This is well solved in PcGive. When selecting
data which enter in VAR ( which will be the basis to determine the number of
cointegration rank for future VECM) you can choose if constant and trend
will be unrestricted or restricted i.e. exogenous. This means that you can
enter (in a VAR model) constant as unrestricted but also as exogenous. This
stays for trend too.
So, you are getting Johansen cases:
Const, Trend - all as restricted
Const - unrestricted, Trend - restricted
Const - restricted, Trend - unrestricted etc.
After such estimation in which you can determine the constant and trend to
be unrestricted or not you procede with usual VAR model and usual testings.
This results are then applyed for VECM estimation.
I don't understand. Allow me some LaTeX-speak to be clearer; assume your VECM
can be written this way:
\Delta y_t = \mu + \alpha \beta' y_{t-1} + \epsilon
where y_t has n rows and r is the cointegration rank, so \alpha and \beta are
(nxr) matrices.
When you restrict eg the constant in a vecm, you force it into the space
spanned by the columns of \alpha. In other words, you hypothesize that the
(nx1) vector \mu can be written as \alpha \cdot m, where m is an unrestricted
(rx1) vector. As a consequence, you're imposing (n-r) constraints. In an
unrestricted VAR, r=n; I fail to see how you can restrict the constant in an
otherwise unrestricted VAR.
Maybe my poor understanding of cointegration is showing? :-)
--
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona
11:22 p.m.
New subject: Cointegration with Gretl -Suggestions!
Riccardo Jack Lucchetti schrieb:
On Wed, February 8, 2006 22:15, john w wrote:
>
>After such estimation in which you can determine the constant and trend to
>be unrestricted or not you procede with usual VAR model and usual testings.
>This results are then applyed for VECM estimation.
There's a misunderstanding there. I think this is the wrong list to
discuss the workings of PcGive, but gretl and PcGive do the same VAR
estimation, which is OLS after all. If you add a trend term in PcGive in
a levels VAR, you implicitly allow for it to cumulate to a quadratic
trend in levels (given a unit root), no matter whether you declare it
"restricted" or not. IIRC that setting is only used for certain stuff,
one of which is VECM.
Maybe my poor understanding of cointegration is showing? :-)
Now that sounds like fishing for compliments ;-)
-sven
7:29 a.m.
New subject: Cointegration with Gretl -Suggestions!
On Thu, February 9, 2006 00:22, Sven Schreiber wrote:
Riccardo Jack Lucchetti schrieb:
> On Wed, February 8, 2006 22:15, john w wrote:
>>
>>After such estimation in which you can determine the constant and trend to
>>be unrestricted or not you procede with usual VAR model and usual testings.
>>This results are then applyed for VECM estimation.
There's a misunderstanding there. I think this is the wrong list to
discuss the workings of PcGive, but gretl and PcGive do the same VAR
estimation, which is OLS after all. If you add a trend term in PcGive in
a levels VAR, you implicitly allow for it to cumulate to a quadratic
trend in levels (given a unit root), no matter whether you declare it
"restricted" or not. IIRC that setting is only used for certain stuff,
one of which is VECM.
I see. I don't have PcGive, so forgive my misunderstanding.
>
> Maybe my poor understanding of cointegration is showing? :-)
>
Now that sounds like fishing for compliments ;-)
*blushes*
--
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona
7:45 a.m.
New subject: Cointegration with Gretl -Suggestions!
There's a misunderstanding there. I think this is the wrong list
to
discuss the workings of PcGive, but gretl and PcGive do the same VAR
estimation, which is OLS after all. If you add a trend term in PcGive in
a levels VAR, you implicitly allow for it to cumulate to a quadratic
trend in levels (given a unit root), no matter whether you declare it
"restricted" or not. IIRC that setting is only used for certain stuff,
one of which is VECM.
You are wright Sven, this is not a list to discuss working of PcGive, but is
a list to discuss gretl's improovement. Comparing is a good way to express
new ideas and features. I mentioned PcGive just to give an example that this
is possible.
Anyway, what do you say about this feature in Gretl?
This will in a good way solve the problem of lags when testing for
cointegration rank.
I will give a look how it technically works.
_________________________________________________________________
Are you using the latest version of MSN Messenger? Download MSN Messenger
7.5 today! http://messenger.msn.co.uk
7:55 a.m.
New subject: Cointegration with Gretl -Suggestions!
Jack:
I see. I don't have PcGive, so forgive my misunderstanding.
No problem Jack.
Sometimes is difficult to express thoughts in just few sentences...I'm get
excited when talking about gretl..and maybe I don't explain things
wright...........:))))
This is the best way I have ever seen to determine the cointegration rank.
This will be absolutely perfect to have it in Gretl if it's possible.
_________________________________________________________________
Are you using the latest version of MSN Messenger? Download MSN Messenger
7.5 today! http://messenger.msn.co.uk
10:05 a.m.
New subject: Cointegration with Gretl -Suggestions!
john w schrieb:
> Jack:
> I see. I don't have PcGive, so forgive my misunderstanding.
No problem Jack.
Sometimes is difficult to express thoughts in just few sentences...I'm
get excited when talking about gretl..and maybe I don't explain things
wright...........:))))
This is the best way I have ever seen to determine the cointegration rank.
This will be absolutely perfect to have it in Gretl if it's possible.
Sorry I still don't get your point; there's no magic going on in pcgive
that would solve all VECM specification problems of gretl users. I have
the impression that you think pcgive is doing something that it's not
really doing. (sentence sounding like greenspan...:-)
-sven
10:36 a.m.
New subject: Cointegration with Gretl -Suggestions!
As I know ( and maybe I'm wrong) these are the steps when determing the
cointegration rank with PcGive.
1. Specify variables - in this step you can choose const and trend to be
unrestricted or not.
2. Estimate VAR model
3. Test estimated VAR model. If model is miss-specified you can add more
lags etc., etc.
4. Run Johansen cointegration test to determine rank.
5. Re-estimating VAR in VECM.
As we discussed yesterday, the main problem in determing cointegration rank
( for example in gretl) is to determin the wright number of lags to be used.
Now, I'm not talking about VECM, I'm talking about testing cointegration
rank which is one step before VECM.
As you know, when determing cointegration rank with gretl you are asked
about lags. Number of lags are closely related to cointegration rank, and
cointegartion rank is very sensitive to entered lags.
Now, tell me on which basis do you enter lags in gretl when testing for
Johansen cointegration ?
Is it AIC or BIC criteria ? If so, who can guarantee you that your model is
not miss-specified.
Do you look at the individual equatations results in gretl cointegration
test output ?
The true is that PcGive gives some more output then other software. There is
no magic.
My point is that gretl becomes better.
Otherways, I will not write so much.
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] Cointegration with Gretl -Suggestions!
Date: Thu, 09 Feb 2006 11:05:05 +0100
john w schrieb:
>> Jack:
>> I see. I don't have PcGive, so forgive my misunderstanding.
>
> No problem Jack.
> Sometimes is difficult to express thoughts in just few sentences...I'm
> get excited when talking about gretl..and maybe I don't explain things
> wright...........:))))
> This is the best way I have ever seen to determine the cointegration
rank.
> This will be absolutely perfect to have it in Gretl if it's possible.
>
Sorry I still don't get your point; there's no magic going on in pcgive
that would solve all VECM specification problems of gretl users. I have
the impression that you think pcgive is doing something that it's not
really doing. (sentence sounding like greenspan...:-)
-sven
_______________________________________________
Gretl-users mailing list
Gretl-users(a)ricardo.ecn.wfu.edu
http://ricardo.ecn.wfu.edu/mailman/listinfo/gretl-users
_________________________________________________________________
The new MSN Search Toolbar now includes Desktop search!
http://toolbar.msn.co.uk/
12:37 p.m.
New subject: Cointegration with Gretl -Suggestions!
On Thu, February 9, 2006 11:36, john w wrote:
As I know ( and maybe I'm wrong) these are the steps when
determing the
cointegration rank with PcGive.
1. Specify variables - in this step you can choose const and trend to be
unrestricted or not.
2. Estimate VAR model
3. Test estimated VAR model. If model is miss-specified you can add more
lags etc., etc.
4. Run Johansen cointegration test to determine rank.
5. Re-estimating VAR in VECM.
In my opinion, this is not very different from what I proposed in my earlier
message
(http://ricardo.ecn.wfu.edu/pipermail/gretl-users/2006-February/000481.html)
As we discussed yesterday, the main problem in determing
cointegration rank
( for example in gretl) is to determin the wright number of lags to be used.
Now, I'm not talking about VECM, I'm talking about testing cointegration
rank which is one step before VECM.
As you know, when determing cointegration rank with gretl you are asked
about lags. Number of lags are closely related to cointegration rank, and
cointegartion rank is very sensitive to entered lags.
Now, tell me on which basis do you enter lags in gretl when testing for
Johansen cointegration ?
The issue looks very simple to me: choose a lag length that YOU consider
suitable on the basis of the VAR estimates and your a-priori opinions (hell,
I'm speaking like a Bayesian here! God forbid!) and then use that lag length
in the Johansen test.
Is it AIC or BIC criteria ? If so, who can guarantee you that your
model is
not miss-specified.
Nobody. Or, better still, any empirical model is misspecified anyway. A model
is a toy world. All you can ask is that your toy is a cool one.
Do you look at the individual equatations results in gretl
cointegration
test output ?
The true is that PcGive gives some more output then other software. There is
no magic.
My point is that gretl becomes better.
Otherways, I will not write so much.
We all appreciate that. But IMO, gretl already does what you need. The fact
that some choices that YOU have to make are not automatically made on your
behalf is a virtue to me, not a defect.
--
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona
12:56 p.m.
New subject: Cointegration with Gretl -Suggestions!
El Jueves, 9 de Febrero de 2006 13:37, Riccardo Jack Lucchetti escribió:
>...
> Is it AIC or BIC criteria ? If so, who can guarantee you that your model
> is not miss-specified.
Nobody. Or, better still, any empirical model is misspecified anyway. A
model is a toy world. All you can ask is that your toy is a cool one.
> Do you look at the individual equatations results in gretl cointegration
> test output ?
>
> The true is that PcGive gives some more output then other software. There
> is no magic.
> My point is that gretl becomes better.
> Otherways, I will not write so much.
We all appreciate that. But IMO, gretl already does what you need. The fact
that some choices that YOU have to make are not automatically made on your
behalf is a virtue to me, not a defect.
I agree completely on that, but ... sometimes may be useful a procedure as
"lagselect.src" (RATS).
There is a very good discussion on selecting the lag order of a VAR in H.
Lütkephol's "Introduction to Multiple Time Series Analysis" (Chapter 4).
--
Ignacio Díaz-Emparanza
Dpto. de Economía Aplicada III (Econometría y Estadística)
UPV-EHU
1:29 p.m.
New subject: Cointegration with Gretl -Suggestions!
On Thu, February 9, 2006 11:36, john w wrote:
> As I know ( and maybe I'm wrong) these are the steps when determing the
> cointegration rank with PcGive.
> 1. Specify variables - in this step you can choose const and trend to be
> unrestricted or not.
> 2. Estimate VAR model
> 3. Test estimated VAR model. If model is miss-specified you can add more
> lags etc., etc.
> 4. Run Johansen cointegration test to determine rank.
> 5. Re-estimating VAR in VECM.
In my opinion, this is not very different from what I proposed in my
earlier
message
(http://ricardo.ecn.wfu.edu/pipermail/gretl-users/2006-February/000481.html)
It's not. This up was for Sven, I didn't know ih he saw our yesterday's
discussion.
It is the the same procedure as you wrote, but it can't cover all Johansen
cases.
I will check sohething this afternoon and will let you know.
The issue looks very simple to me: choose a lag length that YOU
consider
suitable on the basis of the VAR estimates and your a-priori opinions
(hell,
I'm speaking like a Bayesian here! God forbid!) and then use that lag
length
in the Johansen test.
In fact, this is what I'm doing.
Nobody. Or, better still, any empirical model is misspecified anyway.
A
model
is a toy world. All you can ask is that your toy is a cool one.
I agree with you. All models are wrong, but some of them are useful! ;)))
We all appreciate that. But IMO, gretl already does what you need.
And I'm very very greatefull for that.
As I said before I will check something and let you know.
Maybe I raise so much dust for something which is already
implemented......;;))))
_________________________________________________________________
Be the first to hear what's new at MSN - sign up to our free newsletters!
http://www.msn.co.uk/newsletters
7:39 p.m.
New subject: Cointegration with Gretl -Suggestions!
I wrote:
It is the the same procedure as you wrote, but it can't cover all
Johansen
cases.
I will check sohething this afternoon and will let you know.
It can cover all Johansen cases.
Dear all,
my appologies to you. I missed something in comparing estimations.
The PROCEDURE between this two softwares (gretl and PcGive) when testing for
cointegration is simillar, but not the same. In PcGive (as this also
mentioned Sven) you are allowed to select trend as unrestricted which is
later necessary for testing the numbers of cointegration vectors. In PcGive
testing for cointegration vectors and estimating VAR models is in the same
module.
This is not necessary in Gretl as it has this two modules separed.
The truth is that PcGive has more tests, but as Jack said Gretl has
everything you need.
I will add something: tests -> never enough.
Once again sorry for this confusion.
_________________________________________________________________
The new MSN Search Toolbar now includes Desktop search!
http://toolbar.msn.co.uk/
11:58 a.m.
New subject: Feature missing ?
In new gretl version 1.5.0 CVS (02 february 2006) I think that one feature
is missing.
In File module, Open data, after Import function before were listed last
used files.
Now when exiting and then reopening gretl there is no more that list so
everytime when reopening gretl you need to import file.
_________________________________________________________________
The new MSN Search Toolbar now includes Desktop search!
http://toolbar.msn.co.uk/
2:18 p.m.
New subject: Feature missing ?
On Wed, 15 Feb 2006, john w wrote:
In new gretl version 1.5.0 CVS (02 february 2006) I think that one
feature is missing. In File module, Open data, after Import
function before were listed last used files. Now when exiting and
then reopening gretl there is no more that list so everytime when
reopening gretl you need to import file.
OK, I need to check that. I've been working in that general area,
trying to get the opening of files with non-ASCII characters in
their names working correctly on Windows. It looks as if I must
have broken something in the process.
Allin Cottrell
6962
days inactive
6970
days old
18 comments
5 participants
participants (5)
-
Allin Cottrell -
Ignacio Díaz-Emparanza -
john w -
Riccardo Jack Lucchetti -
Sven Schreiber