3:01 p.m.
Hi,
I've just noticed that it is possible to add (in the GUI) lags of the
binary dependent variable in the built-in Probit routine. However, there
are some subtleties involved in this context, so I'm wondering whether
this has always been possible? Could somebody provide some background on
gretl's implementation?
For example, there is this quite recent paper:
ROBERT M. DE JONG, TIEMEN WOUTERSEN
DYNAMIC TIME SERIES BINARY CHOICE
Econometric Theory, 27, 2011, 673–702.
doi:10.1017/S0266466610000472
From the abstract: "... it shows in a time series setting the validity
of the dynamic probit likelihood procedure when lags of the dependent
binary variable are used as regressors ..."
Does gretl rely on this particular paper's findings?
thanks,
sven
3:57 p.m.
On Thu, 26 May 2016, Sven Schreiber wrote:
Hi,
I've just noticed that it is possible to add (in the GUI) lags of the binary
dependent variable in the built-in Probit routine. However, there are some
subtleties involved in this context, so I'm wondering whether this has always
been possible? Could somebody provide some background on gretl's
implementation?
For example, there is this quite recent paper:
ROBERT M. DE JONG, TIEMEN WOUTERSEN
DYNAMIC TIME SERIES BINARY CHOICE
Econometric Theory, 27, 2011, 673–702.
doi:10.1017/S0266466610000472
From the abstract: "... it shows in a time series setting the validity of
the dynamic probit likelihood procedure when lags of the dependent binary
variable are used as regressors ..."
Does gretl rely on this particular paper's findings?
Well, in a way.
Section 3 states quite clearly that maximising the standard log-likelihood
gives you perfctly standard inference. Most of the paper is concerned with
the MSCORE estimator (that we don't have) in a dynamic setting.
It is true that what happens if you stick the --robust option hasn't been
thought out particularly thoroughly. From De Jong and Woutersen (2013):
"Note that given the weak dependence property of Theorem 2, it is also
possible to set forth conditions such that for weakly dependent u_n with
arbitrary distribution,
N^1/2 (b − ) −→ N (0J)
for some matrix J and a β∗ thatuniquely minimizes the objective function.
Here of course β∗ does notnecessarily equal the true parameter value β."
But of course we're sailing the treacherous QMLE seas here, so I doubt
that anybody really knows what happens when :)
-------------------------------------------------------
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
-------------------------------------------------------
7 p.m.
Am 26.05.2016 um 17:57 schrieb Riccardo (Jack) Lucchetti:
On Thu, 26 May 2016, Sven Schreiber wrote:
>
> From the abstract: "... it shows in a time series setting the
> validity of the dynamic probit likelihood procedure when lags of the
> dependent binary variable are used as regressors ..."
>
> Does gretl rely on this particular paper's findings?
Well, in a way.
Section 3 states quite clearly that maximising the standard
log-likelihood gives you perfctly standard inference.
Right. Follow-up: What about h-step forecasting? Does gretl do it in
this case (I know, I should just try for myself...) and if so, does it
do the right thing? AFAI remember the forecasting was one of the main
topics of Kauppi&Saikkonen.
thanks,
sven
9:11 p.m.
On Thu, 26 May 2016, Sven Schreiber wrote:
Am 26.05.2016 um 17:57 schrieb Riccardo (Jack) Lucchetti:
> On Thu, 26 May 2016, Sven Schreiber wrote:
>
>>
>> From the abstract: "... it shows in a time series setting the
>> validity of the dynamic probit likelihood procedure when lags of the
>> dependent binary variable are used as regressors ..."
>>
>> Does gretl rely on this particular paper's findings?
>
> Well, in a way.
>
> Section 3 states quite clearly that maximising the standard
> log-likelihood gives you perfctly standard inference.
Right. Follow-up: What about h-step forecasting? Does gretl do it in this
case (I know, I should just try for myself...) and if so, does it do the
right thing? AFAI remember the forecasting was one of the main topics of
Kauppi&Saikkonen.
I don't know, I haven't read that one. You mean ReStat 2008, right? I'll
read it asap.
-------------------------------------------------------
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
-------------------------------------------------------
10:31 p.m.
A while ago, Sven, I started replicating Thomas&Christian's recession probability
forecasting exercise using gretl's probit environment and lags of the dependent. It
works quite well actually.
If I remember correctly the Kauppi/Saikonnen framework uses the lagged estimated
probabilites instead of the observable, right?
Artur
Am 26. Mai 2016 23:11:50 MESZ, schrieb "Riccardo (Jack) Lucchetti"
<r.lucchetti(a)univpm.it>:
On Thu, 26 May 2016, Sven Schreiber wrote:
> Am 26.05.2016 um 17:57 schrieb Riccardo (Jack) Lucchetti:
>> On Thu, 26 May 2016, Sven Schreiber wrote:
>>
>>>
>>> From the abstract: "... it shows in a time series setting the
>>> validity of the dynamic probit likelihood procedure when lags of
the
>>> dependent binary variable are used as regressors ..."
>>>
>>> Does gretl rely on this particular paper's findings?
>>
>> Well, in a way.
>>
>> Section 3 states quite clearly that maximising the standard
>> log-likelihood gives you perfctly standard inference.
>
> Right. Follow-up: What about h-step forecasting? Does gretl do it in
this
> case (I know, I should just try for myself...) and if so, does it do
the
> right thing? AFAI remember the forecasting was one of the main topics
of
> Kauppi&Saikkonen.
I don't know, I haven't read that one. You mean ReStat 2008, right?
I'll
read it asap.
-------------------------------------------------------
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-devel mailing list
Gretl-devel(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-devel
--
Diese Nachricht wurde von meinem Android-Mobiltelefon mit K-9 Mail gesendet.
3131
days inactive
3131
days old
4 comments
3 participants
participants (3)
-
Artur Tarassow -
Riccardo (Jack) Lucchetti -
Sven Schreiber