5:01 a.m.
Dear Gretl Community,
I trying to estimate a time-varying parameter model (TVP) using the Kalman
filter but I'm getting no success :( The model is a TVP Phillips curve for
Brazilian inflation augmented with the trade openness (my sample has 103
observations):
inf_f = b1(t)*inf(-1) + b2(t)*exp12m + b3(t)*gap(-1) + b4(t)*trade(-1) +
w(t)
b1(t) = b1(t-1) + v1(t)
b2(t) = b2(t-1) + v2(t)
b3(t) = b3(t-1) + v3(t)
b4(t) = b4(t-1) + v4(t)
Where:
inf_f - non monitored prices
exp12m - 12 month expected inflation
gap - output gap
trade - trade openness
Based upon equations (26.1) and (26.2) from the Gretl User's Guide (p. 216)
I'm trying to setup the filter. Please take a look at the following
commands:
list regressors = inf exp12m gap trade
matrix y = {inf_f} # Matrix order:(Txn)=(103x1)
matrix H = *?* # Matrix order:(rxn)=(4x1)
matrix x = {regressors} # Matrix order:(Txk)=(103x4)
matrix F = I(4) # Matrix order:(rxr)=(4x4)
matrix Q = *?* # Matrix order:(rxr)=(4x4)
How can I define H and Q?
Best,
--
*Henrique C. de Andrade*
Doutorando em Economia Aplicada
Universidade Federal do Rio Grande do Sul
www.ufrgs.br/ppge
9:10 a.m.
On Sun, 13 Feb 2011, Henrique Andrade wrote:
Dear Gretl Community,
I trying to estimate a time-varying parameter model (TVP) using the
Kalman filter but I'm getting no success [...]
Henrique,
your state-space model is not correctly specified; in fact, you have no
"regressors": the Phillips curve parameters are your states and the
explanatory variables (output gap etc) form a time-varying H matrix.
I was going to send you an example, but the script I prepared for you
unearthed a bug in CVS gretl, so please leave us a couple of days to fix
this and then I'll send you a (hopefully) working example.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
4:51 p.m.
On Sun, 13 Feb 2011, Riccardo (Jack) Lucchetti wrote:
On Sun, 13 Feb 2011, Henrique Andrade wrote:
> Dear Gretl Community,
> I trying to estimate a time-varying parameter model (TVP) using the Kalman
> filter but I'm getting no success [...]
Henrique,
your state-space model is not correctly specified; in fact, you have no
"regressors": the Phillips curve parameters are your states and the
explanatory variables (output gap etc) form a time-varying H matrix.
I was going to send you an example, but the script I prepared for you
unearthed a bug in CVS gretl, so please leave us a couple of days to fix this
and then I'll send you a (hopefully) working example.
Ok, example attached. It uses some data from Stock & Watson's textbook and
it shouldn't be difficult for you to make sense of it and adapt it to
your needs. Be sure to use the current CVS version to avoid bugs.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
3:19 a.m.
Dear Riccardo,
I'm going to start using your script right now! This work is for my Ph.D.
thesis.
Thanks a lot for your precious help :)
Best regards,
Henrique
2011/2/14 Riccardo (Jack) Lucchetti <r.lucchetti(a)univpm.it>
On Sun, 13 Feb 2011, Riccardo (Jack) Lucchetti wrote:
On Sun, 13 Feb 2011, Henrique Andrade wrote:
>
> Dear Gretl Community,
>> I trying to estimate a time-varying parameter model (TVP) using the
>> Kalman filter but I'm getting no success [...]
>>
>
> Henrique,
> your state-space model is not correctly specified; in fact, you have no
> "regressors": the Phillips curve parameters are your states and the
> explanatory variables (output gap etc) form a time-varying H matrix.
>
> I was going to send you an example, but the script I prepared for you
> unearthed a bug in CVS gretl, so please leave us a couple of days to fix
> this and then I'll send you a (hopefully) working example.
>
Ok, example attached. It uses some data from Stock & Watson's textbook and
it shouldn't be difficult for you to make sense of it and adapt it to your
needs. Be sure to use the current CVS version to avoid bugs.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
--
*Henrique C. de Andrade*
Doutorando em Economia Aplicada
Universidade Federal do Rio Grande do Sul
www.ufrgs.br/ppge
6:56 p.m.
Speaking of time-varying stuff, and in case anyone's interested, attached is a script
that estimates Stock & Watson's unobserved-component stochastic volatility (UC-SV)
model from their 2007 JMCB paper. Also attached is a sample data file with a spot
commodity price index.
It's pretty much a straight translation of Watson's Gauss code (including his
comments), so there are no doubt 'gretl efficiencies' that could be exploited.
Comments welcome!
PS
-----Original Message-----
From: gretl-users-bounces(a)lists.wfu.edu [mailto:gretl-users-bounces@lists.wfu.edu] On
Behalf Of Riccardo (Jack) Lucchetti
Sent: Monday, February 14, 2011 10:51 AM
To: Gretl list
Subject: Re: [Gretl-users] Time Varying Parameters Model
On Sun, 13 Feb 2011, Riccardo (Jack) Lucchetti wrote:
On Sun, 13 Feb 2011, Henrique Andrade wrote:
> Dear Gretl Community,
> I trying to estimate a time-varying parameter model (TVP) using the Kalman
> filter but I'm getting no success [...]
Henrique,
your state-space model is not correctly specified; in fact, you have no
"regressors": the Phillips curve parameters are your states and the
explanatory variables (output gap etc) form a time-varying H matrix.
I was going to send you an example, but the script I prepared for you
unearthed a bug in CVS gretl, so please leave us a couple of days to fix this
and then I'll send you a (hopefully) working example.
Ok, example attached. It uses some data from Stock & Watson's textbook and
it shouldn't be difficult for you to make sense of it and adapt it to
your needs. Be sure to use the current CVS version to avoid bugs.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
4:08 a.m.
Em 14 de fevereiro de 2011 Riccardo (Jack) Lucchetti <r.lucchetti(a)univpm.it>
escreveu:
On Sun, 13 Feb 2011, Riccardo (Jack) Lucchetti wrote:
On Sun, 13 Feb 2011, Henrique Andrade wrote:
>
> Dear Gretl Community,
>> I trying to estimate a time-varying parameter model (TVP) using the
>> Kalman filter but I'm getting no success [...]
>>
>
> (...)
>
Ok, example attached. It uses some data from Stock & Watson's textbook and
it shouldn't be difficult for you to make sense of it and adapt it to your
needs. Be sure to use the current CVS version to avoid bugs.
Dear Jack, now I think I set up my problem using your script. But some
(silly) questions remain:
(1) The estimated coefficients are filtered or smoothed?
(2) How could I retrieve the residuals (to make model diagnosis)?
Once again I would like to thank you so much for your help (this task is
part of my Ph.D. thesis).
Um abraço,
--
*Henrique C. de Andrade*
Doutorando em Economia Aplicada
Universidade Federal do Rio Grande do Sul
www.ufrgs.br/ppge
7:11 a.m.
Em 16 de fevereiro de 2010 Henrique escreveu:
Dear Jack, now I think I set up my problem using your script. But some
(silly) questions remain:
(1) The estimated coefficients are filtered or smoothed?
(2) How could I retrieve the residuals (to make model diagnosis)?
I think I'd found the answer for the first question. Looking at the last
loop of the script (line 104) we can see the user defined function "smpar"
(defined on the lines 45-74) that gives smoothed estimations. I suppose
"smpar" states for "smoothed parameters", am I right Riccardo?
Now I need to modify Riccardo's script in order to make it give me filtered
values (probably applying the kfilter() function).
Once the script starts to give filtered estimates I think I can answer my
second question: How could I retrieve the residuals to make model diagnosis
(correlogram, cusum, normality, etc.).
Looking at Gretl Command Reference I'd found the options to retrieve
information I need: (1)
&E gets the matrix of one-step ahead prediction errors; (2) &V gets the
variance matrix for these errors; (3) &S gets the matrix of estimated values
of the state vector; (4) &P the variance matrix of these estimates and;
(5) &G gets the Kalman gain.
So to continue my work I need find a way to get the filtered parameters and
get the &E, &V, &S, and &P.
Best regards,
--
*Henrique C. de Andrade*
Doutorando em Economia Aplicada
Universidade Federal do Rio Grande do Sul
www.ufrgs.br/ppge
4820
days inactive
4826
days old
6 comments
3 participants
participants (3)
-
Henrique Andrade -
Riccardo (Jack) Lucchetti -
Summers, Peter