11:27 a.m.
Hi! I would like to know what unit is on the left y-axis in the graphs
containing impulse-resonse functions, calculated using bootstrapping from
VEC models in Gretl. Can someone please help me with this?
Best regards,
Lars Ahnland
1:14 p.m.
Em 17 de agosto de 2017, lasses skola <lassesskola(a)gmail.com>:
Hi! I would like to know what unit is on the left y-axis in the graphs
containing impulse-resonse functions, calculated using bootstrapping
from
VEC models in Gretl. Can someone please help me with this?
Dear Lars,
The unit on the left y-axis is standard deviation of the target variable.
Best,
Henrique
1:26 p.m.
Am 17.08.2017 um 13:27 schrieb lasses skola:
Hi! I would like to know what unit is on the left y-axis in the
graphs
containing impulse-resonse functions, calculated using bootstrapping
from VEC models in Gretl. Can someone please help me with this?
Hi, I'm assuming that you are using the SVAR add-on, and that you are
using the scripting interface, otherwise I don't see how you could get
bootstrapping and VEC models together.
(Disclosure: I'm the second author of the SVAR add-on, but the bulk of
that and also of the documentation was written by Jack.)
So, quoting from the documentation pdf file (section 2.4):
"A word on the unit of measurement of IRFs: by their definition (see
equation (7)), and the fact the structural shocks are assumed to have
unit variance, clearly their unit of measurement is the same as the one
for the corresponding observable variable y i,t."
If you have a levels variable in billion dollars, the y-axis unit will
be billion dollars. If you have a growth rate in percentages, so will be
that y-axis.
Perhaps your question is about the underlying shock definition. The IRF
graphs here should be giving you the response of the respective variable
to a unit structural shock. (People, correct me if I'm wrong, it's been
a while since I looked at that add-on.) Sometimes people want to assume
a different structural shock size, e.g. the size of one standard
deviation of the i-th endogenous variable. This could be achieved by
multiplying the IRFs with that SD number.
(The doc also mentions a different normalization which really changes
the y-axis units, if you're interested.)
HTH,
Sven
10:51 a.m.
Hi, and thank you for your response! I am not using any addons to my
knowledge. I was a big vague maybe. There is an option to calculate
confidence intervals of the IRFs using bootstrapping.
I think I put the question in the wrong way as well. What unit is the shock
in? One standard deviation? Or SE of regression?
Yours,
Lars
2017-08-17 15:26 GMT+02:00 Sven Schreiber <svetosch(a)gmx.net>:
Am 17.08.2017 um 13:27 schrieb lasses skola:
> Hi! I would like to know what unit is on the left y-axis in the graphs
> containing impulse-resonse functions, calculated using bootstrapping from
> VEC models in Gretl. Can someone please help me with this?
>
Hi, I'm assuming that you are using the SVAR add-on, and that you are
using the scripting interface, otherwise I don't see how you could get
bootstrapping and VEC models together.
(Disclosure: I'm the second author of the SVAR add-on, but the bulk of
that and also of the documentation was written by Jack.)
So, quoting from the documentation pdf file (section 2.4):
"A word on the unit of measurement of IRFs: by their definition (see
equation (7)), and the fact the structural shocks are assumed to have unit
variance, clearly their unit of measurement is the same as the one for the
corresponding observable variable y i,t."
If you have a levels variable in billion dollars, the y-axis unit will be
billion dollars. If you have a growth rate in percentages, so will be that
y-axis.
Perhaps your question is about the underlying shock definition. The IRF
graphs here should be giving you the response of the respective variable to
a unit structural shock. (People, correct me if I'm wrong, it's been a
while since I looked at that add-on.) Sometimes people want to assume a
different structural shock size, e.g. the size of one standard deviation of
the i-th endogenous variable. This could be achieved by multiplying the
IRFs with that SD number.
(The doc also mentions a different normalization which really changes the
y-axis units, if you're interested.)
HTH,
Sven
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
10:54 a.m.
Am 18.08.2017 um 12:51 schrieb lasses skola:
Hi, and thank you for your response! I am not using any addons to my
knowledge. I was a big vague maybe. There is an option to calculate
confidence intervals of the IRFs using bootstrapping.
OK, can you describe in more detail what you are doing and how?
I think I put the question in the wrong way as well. What unit is the
shock in? One standard deviation? Or SE of regression?
One unit of the respective unit of measurement of the variable, as I
wrote before.
cheers,
sven
1:16 p.m.
On Fri, 18 Aug 2017, lasses skola wrote:
Hi, and thank you for your response! I am not using any addons to my
knowledge. I was a big vague maybe. There is an option to calculate
confidence intervals of the IRFs using bootstrapping.
I think I put the question in the wrong way as well. What unit is the shock
in? One standard deviation? Or SE of regression?
You've had a reply from Sven, but I believe he's thinking about the
SVAR addon rather than the behavior of the built-in "var" and "vecm"
commands.
For the built-in commands, the shock is one standard deviation of the
innovations of the variable in question. Here's a trivial example
(using "var" rather than "vecm" but that doesn't make a
substantive
difference):
<hansl>
open data9-7
var 4 UNEMP PRIME --quiet
matrix IRF = irf(1,1)
print IRF
matrix S = $sigma # covariance matrix of innovations
printf "std deviation of UNEMP innovation: %.5f\n", sqrt(S[1,1])
</hansl>
If you run the above you should see that the IRF vector begins:
0.24541
0.29007
0.27650 ...
while the printf on the last line shows
"std deviation of UNEMP innovation: 0.24541"
hence confirming that the shock is one standard deviation (and the
response is in the "natural" units of the target variable).
Allin Cottrell
1:27 p.m.
Ok, thank you for clarification. I have another question then: Since Gretl
has an option to bootstrap the confidence interval, does that mean that the
impulse-response functions are valid even though the residuals of each VEC
equation are non-normal it their distribution? I know the t- and p-values
of the VEC output are unreliable with non-normal residual distribution, but
doesn't bootstrapping remedy this, since it relies on the "true probability
distribution"? Forgive me if I got it totally wrong, I'm only a beginner at
this.
Best regards,
Lars
2017-08-18 15:16 GMT+02:00 Allin Cottrell <cottrell(a)wfu.edu>:
On Fri, 18 Aug 2017, lasses skola wrote:
Hi, and thank you for your response! I am not using any addons to my
> knowledge. I was a big vague maybe. There is an option to calculate
> confidence intervals of the IRFs using bootstrapping.
> I think I put the question in the wrong way as well. What unit is the
> shock
> in? One standard deviation? Or SE of regression?
>
You've had a reply from Sven, but I believe he's thinking about the SVAR
addon rather than the behavior of the built-in "var" and "vecm"
commands.
For the built-in commands, the shock is one standard deviation of the
innovations of the variable in question. Here's a trivial example (using
"var" rather than "vecm" but that doesn't make a substantive
difference):
<hansl>
open data9-7
var 4 UNEMP PRIME --quiet
matrix IRF = irf(1,1)
print IRF
matrix S = $sigma # covariance matrix of innovations
printf "std deviation of UNEMP innovation: %.5f\n", sqrt(S[1,1])
</hansl>
If you run the above you should see that the IRF vector begins:
0.24541
0.29007
0.27650 ...
while the printf on the last line shows
"std deviation of UNEMP innovation: 0.24541"
hence confirming that the shock is one standard deviation (and the
response is in the "natural" units of the target variable).
Allin Cottrell
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
2:46 p.m.
Am 18.08.2017 um 15:16 schrieb Allin Cottrell:
On Fri, 18 Aug 2017, lasses skola wrote:
> Hi, and thank you for your response! I am not using any addons to my
> knowledge. I was a big vague maybe. There is an option to calculate
> confidence intervals of the IRFs using bootstrapping.
You're absolutely right, I forgot and then overlooked this option.
> I think I put the question in the wrong way as well. What unit is
the
> shock
> in? One standard deviation? Or SE of regression?
You've had a reply from Sven, but I believe he's thinking about the SVAR
addon rather than the behavior of the built-in "var" and "vecm"
commands.
OK, two or three things here:
- If you click on the "help" button in the impulse response window you
get a text about the bootstrap, but the answer to the shock size
question is not there, it is only in the doc for the irf() function ("an
impulse of one standard deviation in the shock variable"). Maybe this
could be copied to the GUI help text.
- I just tried your (Allin's) example setup with UNEMP PRIME in the SVAR
addon, and actually I get a very similar immediate effect of
UNEMP->UNEMP. So I think I was wrong when I inferred from the SVAR doc
that the shock has unit size. (There is a slight difference, but I think
that's just the usual issue of d-o-f corrections or not in the sigma
estimate.) Perhaps the docs for the SVAR add-on need to be clarified
(note to self); or Jack could weigh in on this.
- Finally a question about the irf() function in a scripting context: If
I want to call it several times for several different variables and/or
shocks, together with bootstrapping, does it mean the bootstrap is rerun
every time? That would seem quite inefficient. (In contrast, in the GUI
one can request all responses at once.)
thanks,
sven
4:39 p.m.
On Fri, 18 Aug 2017, Sven Schreiber wrote:
- I just tried your (Allin's) example setup with UNEMP PRIME in
the SVAR
addon, and actually I get a very similar immediate effect of UNEMP->UNEMP. So
I think I was wrong when I inferred from the SVAR doc that the shock has unit
size. (There is a slight difference, but I think that's just the usual issue
of d-o-f corrections or not in the sigma estimate.) Perhaps the docs for the
SVAR add-on need to be clarified (note to self); or Jack could weigh in on
this.
I'll try to contribute to the discussion once I'm back from holiday (end
of the month). :-)
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------
4:56 p.m.
On Fri, 18 Aug 2017, Riccardo (Jack) Lucchetti wrote:
On Fri, 18 Aug 2017, Sven Schreiber wrote:
> - I just tried your (Allin's) example setup with UNEMP PRIME in the SVAR
> addon, and actually I get a very similar immediate effect of UNEMP->UNEMP.
> So I think I was wrong when I inferred from the SVAR doc that the shock has
> unit size. (There is a slight difference, but I think that's just the usual
> issue of d-o-f corrections or not in the sigma estimate.) Perhaps the docs
> for the SVAR add-on need to be clarified (note to self); or Jack could
> weigh in on this.
I'll try to contribute to the discussion once I'm back from holiday (end of
the month). :-)
Enjoy your holiday! (Mine's about to expire.)
Allin
5:41 p.m.
So what about my last question? Are IRF confidence intervals based on
bootstrapping valid even if the VECM has non-normal residuals?
fredag 18 augusti 2017 skrev Allin Cottrell <cottrell(a)wfu.edu>:
On Fri, 18 Aug 2017, Riccardo (Jack) Lucchetti wrote:
On Fri, 18 Aug 2017, Sven Schreiber wrote:
>
> - I just tried your (Allin's) example setup with UNEMP PRIME in the SVAR
>> addon, and actually I get a very similar immediate effect of UNEMP->UNEMP.
>> So I think I was wrong when I inferred from the SVAR doc that the shock has
>> unit size. (There is a slight difference, but I think that's just the usual
>> issue of d-o-f corrections or not in the sigma estimate.) Perhaps the docs
>> for the SVAR add-on need to be clarified (note to self); or Jack could
>> weigh in on this.
>>
>
> I'll try to contribute to the discussion once I'm back from holiday (end
> of the month). :-)
>
Enjoy your holiday! (Mine's about to expire.)
Allin
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
11 p.m.
Bootstrapped IRF confidence levels are used both because of problems in the
use of analytic expressions when disturbances are normal and in the
presence of non-normal distributions. To obtain a better understanding you
should consult an appropriate text. There is material in Lutkepohl and
Kratzig (2004), Applied Time Series Econometrics, Cambridge. or Neusser
(2016), Time Series Econometrics, Springer..
John C Frain
3 Aranleigh Park
Rathfarnham
Dublin 14
Ireland
www.tcd.ie/Economics/staff/frainj/home.html
mailto:frainj@tcd.ie
mailto:frainj@gmail.com
On 18 August 2017 at 18:41, lasses skola <lassesskola(a)gmail.com> wrote:
So what about my last question? Are IRF confidence intervals based
on
bootstrapping valid even if the VECM has non-normal residuals?
fredag 18 augusti 2017 skrev Allin Cottrell <cottrell(a)wfu.edu>:
> On Fri, 18 Aug 2017, Riccardo (Jack) Lucchetti wrote:
>
> On Fri, 18 Aug 2017, Sven Schreiber wrote:
>>
>> - I just tried your (Allin's) example setup with UNEMP PRIME in the SVAR
>>> addon, and actually I get a very similar immediate effect of
UNEMP->UNEMP.
>>> So I think I was wrong when I inferred from the SVAR doc that the shock has
>>> unit size. (There is a slight difference, but I think that's just the
usual
>>> issue of d-o-f corrections or not in the sigma estimate.) Perhaps the docs
>>> for the SVAR add-on need to be clarified (note to self); or Jack could
>>> weigh in on this.
>>>
>>
>> I'll try to contribute to the discussion once I'm back from holiday (end
>> of the month). :-)
>>
>
> Enjoy your holiday! (Mine's about to expire.)
>
> Allin
> _______________________________________________
> Gretl-users mailing list
> Gretl-users(a)lists.wfu.edu
> http://lists.wfu.edu/mailman/listinfo/gretl-users
>
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
8:27 a.m.
So the quick answer is yes?
2017-08-19 1:00 GMT+02:00 John C Frain <frainj(a)gmail.com>:
Bootstrapped IRF confidence levels are used both because of problems
in
the use of analytic expressions when disturbances are normal and in the
presence of non-normal distributions. To obtain a better understanding you
should consult an appropriate text. There is material in Lutkepohl and
Kratzig (2004), Applied Time Series Econometrics, Cambridge. or Neusser
(2016), Time Series Econometrics, Springer..
John C Frain
3 Aranleigh Park
Rathfarnham
Dublin 14
Ireland
www.tcd.ie/Economics/staff/frainj/home.html
mailto:frainj@tcd.ie
mailto:frainj@gmail.com
On 18 August 2017 at 18:41, lasses skola <lassesskola(a)gmail.com> wrote:
> So what about my last question? Are IRF confidence intervals based on
> bootstrapping valid even if the VECM has non-normal residuals?
>
>
> fredag 18 augusti 2017 skrev Allin Cottrell <cottrell(a)wfu.edu>:
>
>> On Fri, 18 Aug 2017, Riccardo (Jack) Lucchetti wrote:
>>
>> On Fri, 18 Aug 2017, Sven Schreiber wrote:
>>>
>>> - I just tried your (Allin's) example setup with UNEMP PRIME in the
>>>> SVAR addon, and actually I get a very similar immediate effect of
>>>> UNEMP->UNEMP. So I think I was wrong when I inferred from the SVAR doc
that
>>>> the shock has unit size. (There is a slight difference, but I think
that's
>>>> just the usual issue of d-o-f corrections or not in the sigma estimate.)
>>>> Perhaps the docs for the SVAR add-on need to be clarified (note to
self);
>>>> or Jack could weigh in on this.
>>>>
>>>
>>> I'll try to contribute to the discussion once I'm back from holiday
>>> (end of the month). :-)
>>>
>>
>> Enjoy your holiday! (Mine's about to expire.)
>>
>> Allin
>> _______________________________________________
>> Gretl-users mailing list
>> Gretl-users(a)lists.wfu.edu
>> http://lists.wfu.edu/mailman/listinfo/gretl-users
>>
>
> _______________________________________________
> Gretl-users mailing list
> Gretl-users(a)lists.wfu.edu
> http://lists.wfu.edu/mailman/listinfo/gretl-users
>
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
7:34 p.m.
No: the quick answer is "do some more reading as we'll never know your data
as well as you do", as you've just been politely told.
C
On 19 August 2017 at 09:27, lasses skola <lassesskola(a)gmail.com> wrote:
So the quick answer is yes?
2017-08-19 1:00 GMT+02:00 John C Frain <frainj(a)gmail.com>:
> Bootstrapped IRF confidence levels are used both because of problems in
> the use of analytic expressions when disturbances are normal and in the
> presence of non-normal distributions. To obtain a better understanding you
> should consult an appropriate text. There is material in Lutkepohl and
> Kratzig (2004), Applied Time Series Econometrics, Cambridge. or Neusser
> (2016), Time Series Econometrics, Springer..
>
> John C Frain
> 3 Aranleigh Park
> Rathfarnham
> Dublin 14
> Ireland
> www.tcd.ie/Economics/staff/frainj/home.html
> mailto:frainj@tcd.ie
> mailto:frainj@gmail.com
>
> On 18 August 2017 at 18:41, lasses skola <lassesskola(a)gmail.com> wrote:
>
>> So what about my last question? Are IRF confidence intervals based on
>> bootstrapping valid even if the VECM has non-normal residuals?
>>
>>
>> fredag 18 augusti 2017 skrev Allin Cottrell <cottrell(a)wfu.edu>:
>>
>>> On Fri, 18 Aug 2017, Riccardo (Jack) Lucchetti wrote:
>>>
>>> On Fri, 18 Aug 2017, Sven Schreiber wrote:
>>>>
>>>> - I just tried your (Allin's) example setup with UNEMP PRIME in the
>>>>> SVAR addon, and actually I get a very similar immediate effect of
>>>>> UNEMP->UNEMP. So I think I was wrong when I inferred from the SVAR
doc that
>>>>> the shock has unit size. (There is a slight difference, but I think
that's
>>>>> just the usual issue of d-o-f corrections or not in the sigma
estimate.)
>>>>> Perhaps the docs for the SVAR add-on need to be clarified (note to
self);
>>>>> or Jack could weigh in on this.
>>>>>
>>>>
>>>> I'll try to contribute to the discussion once I'm back from
holiday
>>>> (end of the month). :-)
>>>>
>>>
>>> Enjoy your holiday! (Mine's about to expire.)
>>>
>>> Allin
>>> _______________________________________________
>>> Gretl-users mailing list
>>> Gretl-users(a)lists.wfu.edu
>>> http://lists.wfu.edu/mailman/listinfo/gretl-users
>>>
>>
>> _______________________________________________
>> Gretl-users mailing list
>> Gretl-users(a)lists.wfu.edu
>> http://lists.wfu.edu/mailman/listinfo/gretl-users
>>
>
>
> _______________________________________________
> Gretl-users mailing list
> Gretl-users(a)lists.wfu.edu
> http://lists.wfu.edu/mailman/listinfo/gretl-users
>
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
--
Clive Nicholas
"My colleagues in the social sciences talk a great deal about methodology.
I prefer to call it style." -- Freeman J. Dyson
2459
days inactive
2461
days old
13 comments
7 participants
participants (7)
-
Allin Cottrell -
Clive Nicholas -
Henrique Andrade -
John C Frain -
lasses skola -
Riccardo (Jack) Lucchetti -
Sven Schreiber