10:07 p.m.
Good afternoon.
I am trying to estimate the parameters of a model that also has latent
state variables, using the Kalman filter and maximum likelihood.
After setting up the filter, which I run and it appears to work fine on
its own (i.e treating the parameters fixed to their initial values), I
follow the example script in p. 313 of User's guide, where the mle
command line is
mle logl = ERR ? NA : kb.llt (where "kb" is the name of the Kalman
bundle)
and inside the mle command we see
ERR = kfilter(&kb)
I wrote the same syntax but using "ksmooth" instead of "kfilter".
I get the message
<<
"llt": no such item
The formula 'logl = ERR ? NA : kb.llt'
produced an error on execution
>
I tried also to embed the lot into a function, as the example script
does, but I continued to get the same error message.
Any ideas?
--
Alecos Papadopoulos PhD
Athens University of Economics and Business
web: alecospapadopoulos.wordpress.com/
skype:alecos.papadopoulos
11:40 p.m.
New subject: Kalman and MLE issue
On Mon, 30 Dec 2019, Alecos Papadopoulos wrote:
I am trying to estimate the parameters of a model that also has
latent state
variables, using the Kalman filter and maximum likelihood.
After setting up the filter, which I run and it appears to work fine on its
own (i.e treating the parameters fixed to their initial values), I follow the
example script in p. 313 of User's guide, where the mle command line is
mle logl = ERR ? NA : kb.llt (where "kb" is the name of the Kalman
bundle)
and inside the mle command we see
ERR = kfilter(&kb)
I wrote the same syntax but using "ksmooth" instead of "kfilter".
I get the message
"llt": no such item
The formula 'logl = ERR ? NA : kb.llt'
produced an error on execution
You need a forward pass (kfilter) to get the loglikelihood. The
backward pass (ksmooth) gives you smoothed estimates of the state
and its variance after filtering.
Allin
11:51 p.m.
New subject: Kalman and MLE issue
ok, thanks Allin.
Maybe the following part in the User guide should be rephrased in some
future version, Section 34.6, pp. 308, presenting the ksmooth thing
<<On successful completion, */all the quantities/* computed by kfilter
are available as bundle members
(see section 34.5),...>>
This is what gave me the impression that I could use ksmooth (this and
the first sentence in p. 308 that says <<Note that since ksmooth starts
with a forward pass, it can be run without a prior call to kfilter.>>)
Alecos Papadopoulos PhD
Athens University of Economics and Business
web: alecospapadopoulos.wordpress.com/
skype:alecos.papadopoulos
On 30/12/2019 01:40, Allin Cottrell wrote:
On Mon, 30 Dec 2019, Alecos Papadopoulos wrote:
> I am trying to estimate the parameters of a model that also has
> latent state variables, using the Kalman filter and maximum likelihood.
>
> After setting up the filter, which I run and it appears to work fine
> on its own (i.e treating the parameters fixed to their initial
> values), I follow the example script in p. 313 of User's guide, where
> the mle command line is
>
> mle logl = ERR ? NA : kb.llt (where "kb" is the name of the
> Kalman bundle)
>
> and inside the mle command we see
>
> ERR = kfilter(&kb)
>
> I wrote the same syntax but using "ksmooth" instead of
"kfilter".
>
> I get the message
>
> "llt": no such item
> The formula 'logl = ERR ? NA : kb.llt'
> produced an error on execution
You need a forward pass (kfilter) to get the loglikelihood. The
backward pass (ksmooth) gives you smoothed estimates of the state and
its variance after filtering.
Allin
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website: https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
12:02 a.m.
New subject: Kalman and MLE issue
On Mon, 30 Dec 2019, Alecos Papadopoulos wrote:
ok, thanks Allin.
Maybe the following part in the User guide should be rephrased in some future
version, Section 34.6, pp. 308, presenting the ksmooth thing
<<On successful completion, */all the quantities/* computed by kfilter are
available as bundle members
(see section 34.5),...>>
This is what gave me the impression that I could use ksmooth (this and the
first sentence in p. 308 that says <<Note that since ksmooth starts with a
forward pass, it can be run without a prior call to kfilter.>>)
Good point. Looks like we need to correct something there, either in
the doc or the substance of what's going on.
Jack -- any thoughts? (When you get back!)
Allin
1:15 a.m.
New subject: Kalman and MLE issue
My input as regards functionality here is the following: state-space
modeling and the Kalman filter are often used also in models where the
number of observations are relatively few, and the focus is on
parameter/latent states estimation, rather than forecasting (like
macroeconomoetric models that look to interpret the past). In such a
setup, it may be preferable to obtain estimates based on the whole
realized sample, i.e. use information retrospectively (this is why I
used ksmooth in my model), since there is not much of it. So if there is
a painless way to allow ksmooth to also run with mle, it would be a
useful option.
Thanks again Allin.
Alecos Papadopoulos PhD
Athens University of Economics and Business
web: alecospapadopoulos.wordpress.com/
skype:alecos.papadopoulos
On 30/12/2019 02:02, Allin Cottrell wrote:
> On Mon, 30 Dec 2019, Alecos Papadopoulos wrote:
>
>> ok, thanks Allin.
>>
>> Maybe the following part in the User guide should be rephrased in
>> some future version, Section 34.6, pp. 308, presenting the ksmooth thing
>>
>> <<On successful completion, */all the quantities/* computed by
>> kfilter are available as bundle members
>> (see section 34.5),...>>
>>
>> This is what gave me the impression that I could use ksmooth (this
>> and the first sentence in p. 308 that says <<Note that since ksmooth
>> starts with a forward pass, it can be run without a prior call to
>> kfilter.>>)
>
> Good point. Looks like we need to correct something there, either in
> the doc or the substance of what's going on.
>
> Jack -- any thoughts? (When you get back!)
>
> Allin
> _______________________________________________
> Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
> To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
> Website:
> https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
>
10:57 p.m.
New subject: Kalman and MLE issue
On Sun, 29 Dec 2019, Allin Cottrell wrote:
On Mon, 30 Dec 2019, Alecos Papadopoulos wrote:
> ok, thanks Allin.
>
> Maybe the following part in the User guide should be rephrased in some
> future version, Section 34.6, pp. 308, presenting the ksmooth thing
>
> <<On successful completion, */all the quantities/* computed by kfilter are
> available as bundle members
> (see section 34.5),...>>
>
> This is what gave me the impression that I could use ksmooth (this and the
> first sentence in p. 308 that says <<Note that since ksmooth starts with a
> forward pass, it can be run without a prior call to kfilter.>>)
Good point. Looks like we need to correct something there, either in the doc
or the substance of what's going on.
Jack -- any thoughts? (When you get back!)
I'm just back. :)
I see no good reason why we couldn't offer the likelihood after ksmooth(),
although in principle that's something you only need the forward pass for.
Therefore, inserting ksmooth() in a mle loop will just waste electricity
and release unnecessary CO2 into the atmosphere. Whenever you need to
estimate parameters via MLE and the loglikelihood is computed via the
Kalman filter, the sane thing to do is use kfilter() inside the mle loop
and ksmooth() right afterwards if you need the smoothed states.
If we decide to go for the environment-friendly option and keep things the
way they are now, perhaps we could add the paragraph above to the
documentation.
-------------------------------------------------------
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
-------------------------------------------------------
1:18 a.m.
New subject: Kalman and MLE issue
It appears I was mistakenly under the impression that using the
smoothing method for the states would affect the value of the likelihood
and so in turn the mle estimates. Ah, I need to revisit the theory.
It would certainly be helpful to clarify in the Guide the
kfilter-mle-ksmooth/kdsmooth relations.
Alecos Papadopoulos PhD
Athens University of Economics and Business
web: alecospapadopoulos.wordpress.com/
skype:alecos.papadopoulos
On 7/1/2020 00:57, Riccardo (Jack) Lucchetti wrote:
On Sun, 29 Dec 2019, Allin Cottrell wrote:
> On Mon, 30 Dec 2019, Alecos Papadopoulos wrote:
>
>> ok, thanks Allin.
>>
>> Maybe the following part in the User guide should be rephrased in
>> some future version, Section 34.6, pp. 308, presenting the ksmooth
>> thing
>>
>> <<On successful completion, */all the quantities/* computed by
>> kfilter are available as bundle members
>> (see section 34.5),...>>
>>
>> This is what gave me the impression that I could use ksmooth (this
>> and the first sentence in p. 308 that says <<Note that since ksmooth
>> starts with a forward pass, it can be run without a prior call to
>> kfilter.>>)
>
> Good point. Looks like we need to correct something there, either in
> the doc or the substance of what's going on.
>
> Jack -- any thoughts? (When you get back!)
I'm just back. :)
I see no good reason why we couldn't offer the likelihood after
ksmooth(), although in principle that's something you only need the
forward pass for. Therefore, inserting ksmooth() in a mle loop will
just waste electricity and release unnecessary CO2 into the
atmosphere. Whenever you need to estimate parameters via MLE and the
loglikelihood is computed via the Kalman filter, the sane thing to do
is use kfilter() inside the mle loop and ksmooth() right afterwards if
you need the smoothed states.
If we decide to go for the environment-friendly option and keep things
the way they are now, perhaps we could add the paragraph above to the
documentation.
-------------------------------------------------------
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
-------------------------------------------------------
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website: https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
1793
days inactive
1802
days old
6 comments
3 participants
participants (3)
-
Alecos Papadopoulos -
Allin Cottrell -
Riccardo (Jack) Lucchetti