3:37 p.m.
Hi community!
I'm trying to compute a time-varying state space model to calculate stocks' betas.
Unfortunately, I have some missing observations and, when this is the case, the model
implemented by Gretl suddenly stops and doesn't compute anything (as expected). My
question is simple: Is there a way of instructing Gretl to ignore missing values and
compute the state spacel model?
Here is the example code taken from chapter 35 of Gretl manual. Please note that I added
line 6 "INFQ[20] = NA" in order to have a missing data point in the data.
<hansl>
function void at_each_step(bundle *b)
b.obsymat = transp(b.mX[b.t,])
end function
open AWM.gdt --quiet
smpl 1974:1 1994:1
INFQ[20] = NA # added this line to have a missing value
/* parameter initialization */
scalar b0 = mean(INFQ)
scalar s_obs = 0.1
scalar s_state = 0.1
/* bundle setup */
bundle B = ksetup(INFQ, 1, 1, 1)
matrix B.mX = {URX}
matrix B.depvar = {INFQ}
B.timevar_call = "at_each_step"
B.diffuse = 1
/* ML estimation of intercept and the two variances */
mle LL = err ? NA : B.llt
B.obsy = B.depvar - b0
B.obsvar = s_obs^2
B.statevar = s_state^2
err = kfilter(&B)
params b0 s_obs s_state
end mle
/* display the smoothed time-varying slope */
ksmooth(&B)
series tvar_b1hat = B.state[,1]
series tvar_b1se = sqrt(B.stvar[,1])
gnuplot tvar_b1hat --time-series --with-lines --output=display \
--band=tvar_b1hat,tvar_b1se,1.96 --band-style=fill
</hansl>
Best,
Filipe
12:10 a.m.
Am 21.12.2020 um 16:37 schrieb Filipe Costa:
Hi community!
I'm trying to compute a time-varying state space model to calculate
stocks' betas. Unfortunately, I have some *missing observations* and,
when this is the case, the model implemented by Gretl suddenly stops and
doesn't compute anything (as expected). My question is simple: *Is there
a way of instructing Gretl to ignore missing values and compute the
state spacel model?*
Here is the example code taken from chapter 35 of Gretl manual. Please
note that I added line 6 "*INFQ[20] = NA*" in order to have a missing
data point in the data.
Hi Filipe, I would suggest to take a look at Jack's (Lucchetti) article
in the JSS a few years back, "State Space Methods in Gretl", although
I'm not sure if that is the latest available version.
In some places I think he simply put in a misszero() conversion, but
only he can tell whether that was really the key.
cheers
sven
12:43 a.m.
On Wed, 23 Dec 2020, Sven Schreiber wrote:
Am 21.12.2020 um 16:37 schrieb Filipe Costa:
> Hi community!
>
> I'm trying to compute a time-varying state space model to calculate
> stocks' betas. Unfortunately, I have some *missing observations* and,
> when this is the case, the model implemented by Gretl suddenly stops and
> doesn't compute anything (as expected). My question is simple: *Is there
> a way of instructing Gretl to ignore missing values and compute the
> state spacel model?*
>
> Here is the example code taken from chapter 35 of Gretl manual. Please
> note that I added line 6 "*INFQ[20] = NA*" in order to have a missing
> data point in the data.
Hi Filipe, I would suggest to take a look at Jack's (Lucchetti) article
in the JSS a few years back, "State Space Methods in Gretl", although
I'm not sure if that is the latest available version.
Please don't! :D
That article is very outdated, and uses an old and deprecated syntax.
In some places I think he simply put in a misszero() conversion, but
only he can tell whether that was really the key.
I believe that Filipe's script ought to work. I'm trying to figure out
what's wrong.
-------------------------------------------------------
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
-------------------------------------------------------
1:07 a.m.
On Wed, 23 Dec 2020, Riccardo (Jack) Lucchetti wrote:
> In some places I think he simply put in a misszero() conversion,
but
> only he can tell whether that was really the key.
I believe that Filipe's script ought to work. I'm trying to figure out what's
wrong.
OK, I got it now: Sven is right, the missing entries in the loglikelihood
should be turned into zeros. However, for doing this you need to be able
to turn the "llt" element of the Kalman bundle (which is a matrix) into a
series, and to do that, you need to ensure that they are the same length.
Therefore, to prevent gretl from dropping the rows with missing entries,
you need to set skip_missing to "off", as in (I'm only showing the
relevant part of the script):
<hansl>
set skip_missing off
/* ML estimation of intercept and the two variances */
mle LL = err ? NA : misszero(llik)
B.obsy = B.depvar - b0
B.obsvar = s_obs^2
B.statevar = s_state^2
err = kfilter(&B)
series llik = B.llt
params b0 s_obs s_state
end mle
</hansl>
Alternatively, you can emulate the behaviour of misszero() "by hand",
without using the "set" command, as in
<hansl>
mle LL = err ? NA : llik
B.obsy = B.depvar - b0
B.obsvar = s_obs^2
B.statevar = s_state^2
err = kfilter(&B)
series llik = ok(B.llt) ? B.llt : 0
params b0 s_obs s_state
end mle
</hansl>
Two questions for Allin:
(a) would you object to extending the misszero() function to matrix
arguments?
(b) some time ago, we discussed the possibility of allowing printf
statements inside mle blocks. How do you feel about that?
-------------------------------------------------------
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
-------------------------------------------------------
1542
days inactive
1544
days old
3 comments
3 participants
participants (3)
-
Filipe Costa -
Riccardo (Jack) Lucchetti -
Sven Schreiber