Re: [Gretl-users] Activating HAC does not work
by Allin Cottrell
On Tue, 26 Apr 2011, Riccardo (Jack) Lucchetti wrote:
> On Tue, 26 Apr 2011, Sven Schreiber wrote:
>
> > I tend to think it should be _possible_ to use HAC with VARs for
> > demonstration purposes, even if it may not be wise to use them for real
> > applications.
> >
> > The robust default should probably be the "wise" one, i.e. HC but not
> > HAC. However, there may also be a case to treat all time-series models
> > alike, as you mention.
>
> I agree with Sven. Besides, it is entirely possible that you have
> heteroskedasticity in a well-specified VAR, so some form of adjustment may
> be necessary after all.
Heteroskedasticity is handled OK: --robust works for VARs,
producing one or other variant of HC* depending on the "set"
options. The problem Artur indicated was that you can't activate
Newey-West for VARs, although it's the default for single-equation
OLS models on time-series data (unless "force_hc" is turned on).
I think maybe the simplest and most backward-compatible thing to
do here is (a) leave plain HC as the --robust default for VARs (as
it has been for quite a while now), but (b) introduce a --hac
option specific to VARs that enables Newey-West if you need it.
(Oh, and (c): update the documentation.)
Does this sound OK?
Allin
12 years, 12 months
Re: [Gretl-users] Activating HAC does not work
by Allin Cottrell
On Mon, 25 Apr 2011, Artur Tarassow wrote:
> I am just estimating some VAR models and would like to use robust
> standard errors. I am using the following lines to set up HAC...
You're right, these won't work to produce HAC for a VAR. It's not
exactly a bug, but a semi-deliberate decision ;-)
That is, some time ago we replaced equation-by-equation estimation
of VARs by a matrix method that does the whole thing in one go. At
that time I rebuilt the HC variance estimator for the new method
but I didn't bother rebuilding the HAC estimator. The reason
(other than laziness) was that you'd generally expect a VAR to
include enough lags to make HAC redundant. (Stock and Watson, for
example, include several VARS in their undergraduate textbook and
they always use a robust variance estimator, but they never use
HAC for VARs: I asked them why not, and that's the answer they
gave me.)
Anyway, it's easy enough to re-enable HAC for VARs if anyone
really wants it. But if I do so, what should the default be?
Should VARs be treated like regular models on time-series data
with regard to the --robust option (that is, HAC unless you "set
force_hc on")? Or vice versa (with a new VAR-specific "set"
variable, "force_hac")?
What do people think?
Allin
12 years, 12 months
a somewhat peculiar question
by artur bala
Dear Allin,
I was asked by some colleagues of mine about the praying boy in the
gretl's icon. What does it stand for? Is there any connection with some
statistical issues? Do you think a more, let's say,
econometric-inspired icon will increase the chances of gretl to get noticed?
Best regards,
artur
12 years, 12 months
Re: [Gretl-users] how to estimate a censored or truncated model?
by Allin Cottrell
On Mon, 25 Apr 2011, Qi Shi wrote:
> thanks allin. I manage it. However, i think gretl maybe have too
> little choices for LDV models. Say, in the duration model, why it
> require the users to use log-normal but not normal distribution.
Because durations are inherently non-negative. The normal
distribution is not applicable.
> Moreover, it incorprates censored (and truncated?) models into the
> duration models...
It doesn't. Duration models are just one particular sub-class of
limited dependent variable (LDV) models: the dependent variable is
limited in that it cannot be negative, and it may be
right-censored (if some "spells" are not finished yet).
Other sorts of LDV models are handled by different estimators:
logit, probit (binary, ordered or bivariate), tobit, intreg
(interval regression), poisson and negbin for count data, and so
on.
Allin Cottrell
12 years, 12 months
Activating HAC does not work
by Artur Tarassow
Hi gretl community,
I am just estimating some VAR models and would like to use robust
standard errors. I am using the following lines to set up HAC:
---------------------
set force_hc off #to make sure that HAC and not HC is used
set hac_prewhiten off #Andrews-Monahan prewhitening and re-coloring
#set hac_kernel bartlett #Alternative: Parzen kernel or QS
#set hac_lag nw3 #default when prewhitening is selected
var p Y --robust
---------------------
But still the output uses HC instead of HAC:
----------------
I
Current session: 2011-04-25 17:58
? set force_hc off
? set hac_prewhiten off #Andrews-Monahan prewhitening and re-coloring
#set hac_kernel bartlett #Alternative: Parzen kernel or QS
#set hac_lag nw3
? scalar q = 4
Replaced scalar q = 4
? Mod3 <- var q Y --robust
VAR system, lag order 4
OLS estimates, observations 1955:1-2010:4 (T = 224)
Log-likelihood = 1712.7457
Determinant of covariance matrix = 7.8281352e-10
AIC = -15.1317
BIC = -14.8575
HQC = -15.0210
Portmanteau test: LB(48) = 166.795, df = 176 [0.6786]
Equation 1: ws
Heteroskedasticity-robust standard errors, variant HC1
------------------------
I am not sure whether it's my fault or a bug.
I am using the current cvs on Ubuntu 11.04.
Best,
Artur
12 years, 12 months
Re: [Gretl-users] how to estimate a censored or truncated model?
by Allin Cottrell
On Mon, 25 Apr 2011, Qi Shi wrote:
> thanks Allin, I read this User's Guide, but it does not tell me how to
> use buttons. Say, to do Wooldridge's example 17.4 "Duration of
> Recidivism". This example has a variable "cens" to indicate which
> observations are censored.
Um, in the duration model window there's a button labeled
"Censoring variable" and a Help button that gives a detailed
description. What more do you want?
Allin Cottrell
12 years, 12 months
Re: [Gretl-users] how to estimate a censored or truncated model?
by Allin Cottrell
On Sat, 23 Apr 2011, Qi Shi wrote:
> how to estimate a censored or truncated model? I find no these
> buttons. thanks
Under /Model/Nonlinear models you have various relevant options.
Also see the help for "probit", "tobit", "intreg", and the chapter
of the Gretl User's Guide titled "Discrete and Censored Dependent
Variables".
Allin Cottrell
12 years, 12 months
Strange Behaviour Updating Gretl on Mac OS/X
by Henrique Andrade
Dear Gretl Community,
Recently I noticed that the Gretl updating procedure is a lit bit different.
In the past, all I needed was to drag the file Gretl.app (inside the
gretl-intel.dmg file) to my Applications folder. Now I need firstly to
delete the older version and then copy the new one to Applications folder.
Now, if I try to update in the old way Mac OS/X tells me the file is blocked
and I have no permission to overwrite Gretl.app (I'm the only administrator
of my machine).
Is this an expected behaviour?
Best regards,
--
*Henrique C. de Andrade*
Doutorando em Economia Aplicada
Universidade Federal do Rio Grande do Sul
www.ufrgs.br/ppge
12 years, 12 months
gretl 1.9.5 (Windows and OS X) and gig
by Allin Cottrell
Hello all,
Due to an unfortunate packaging error, the installer packages for
gretl 1.9.5 on MS Windows and OS X went out to sourceforge (on
April 22) containing a test version of the "gig" addon instead of
the most recent "production" version.
gig is GARCH-in-gretl, a sophisticated package for estimating a
wide variety of GARCH models written by Jack Lucchetti.
We have now (April 24) fixed these packages on sourceforge. If you
downloaded gretl 1.9.5 before Sun Apr 24 12:40 EDT, and are
interested in GARCH estimation, we urge you to go back for the
replacement version. It should show a build date of 2011-04-24
under /Help/About.
Sorry for the inconvenience.
--
Allin Cottrell
Department of Economics
Wake Forest University
12 years, 12 months
scatter plot with non-binary factor separation
by Ferenci Tamas
Dear gretl community,
I was just wondering: is it somehow possible to create a scatter plot
where markers depend on a categorical (but not dummy, i.e. not binary)
variable? View / Graph specified vars / X-Y with factor separation
looked promising, but it only allows dummy variables as separating
factors, so the points can only be partitioned into two sets. Is it
somehow possible to extend this to more than two...?
(I thought the answer is surely positive as this is a very-very
logical and obvious feature, but I couldn't figure out how to do this
in gretl...)
Thank you in advance,
Tamas
13 years
