About two months ago when I first started working with Gretl I asked if
Gretl has tests for randomness of model residuals (accessible from
either the GUI or the command line). I got a number of fairly
non-specific responses like "look at the documentation" and "look at the
menus on the windows". Well, after two months of working with Gretl I am
now a hardened veteran and ready to ask the question again:
The only mention of a test for randomness I can find is the
non-parametric runs test. Can Gretl do a Ljung-Box test (or some other
test I am not familiar with) of the randomness of residuals? Thanks in
advance for any advice you may offer.
In the Gretl GUI one can define a new variable by ADD -> DEFINE NEW
VARIABLE. This works quite nicely, and I can define something like
y=cos(2*pi*x/5) and use it in a regression. In addition, I can
right-click on y, select EDIT ATTRIBUTES, and edit the formula, changing
it to y=cos(2*pi*x/10) for example. In general, this is a very useful
feature of Gretl.
However, if I save the file and reload it the "formula" becomes a
"description" in the EDIT ATTRIBUTES window. This means I can no longer
edit the formula. Is there a way to retain the ability to edit the
formula after saving a session?
I'm an undergrad, just started looking at gretl. It looks amazing; thanks a
bunch for the work.
I have two minor suggestions.
1) I noticed you can make a covariance coefficient matrix. Have you
considered including the standard errors under these, and also using the
asterisks to mark whether the coefficients are statistically significant at
some level? Could that be added as an option? That's how Stock and Watson
seem to present these coefficient matrices, and they also include summary
statistics at the bottom (SER, R-bar squared, and n).
2) The output of a least squares regression is somewhat long, and I've
noticed that I can't selectively copy/paste from it. That's not a big deal
as I can delete stuff after copying, but overall I wonder why that is, and
would like to be able to copy/paste selectively. I've noticed that in
general I can't right-click on these outputs, and I wonder why.
3) It would also be nice to just output an equation that I could copy/paste,
or export somehow.
I imagine this is the best open-source stats program around? Can gretl do
time-series well? How does gretl compare to the commercial software (I've
used Eviews 4.1 Student Version and SPSS, and both seem terribly clunky).
Dear gretl users,
Given variable y follows a quadratic trend.
1) In ADF test should be chosen variant with constant and linear trend and this test should be made for the levels of variable (not differences)?
2) I'd like to confirm results obtained from ADF test by doing KPSS test. KPSS test should be made for first differences with chosing linear trend?
3) While testing cointegration (Engle -Granger test) should I chose version with constant (or with constant and linear trend)?
Thanks in advance.
Are heteroskedasticity tests other than the White test available built-in in
Namely, I'm looking for an easy way to carry out the Breusch-Pagan/Godfrey
and the Goldfeld-Quant tests of heteroskedasticity.
(Of course I don't know if these names are the standard names of test, as I
am using the names in our textbook Johnston-DiNardo.)
And I'm sorry if my question is asked before! I did not know how I could
search the mailing list archive.
Amir Reza Khosroshai
Why gretl does not display the const term in the fixed-effects panel model?
The random effects and the pooled OLS models do display const. Is there a
way to show the estimated constant?
I'll probably just try to do an adequate job of it for now (the paper needs to be ready for a conference), but I can redo the analysis before submitting it to a journal.
I think though that I can pretty much do everything I need to do using the arbond estimation (and I've found sources for the data I'll needed), with the exception of the categorical problem--mind you, I can always use the time-honored cop-out of taking the log of the number of deaths.
-------------- Original message --------------
From: Sven Schreiber <svetosch(a)gmx.net>
> Am 13.03.2008 15:47, scottd.orr(a)comcast.net schrieb:
> > One more question: my dependent variable arguably should be
> > categorical. Is there any way to deal with this with gretl's panel
> > routines, and if not, any reasonably doable way I could do this outside
> > of the canned routines?
> uh-oh -- any other fancy cutting-edge stuff you want to do?
> seriously, it is my impression by now that if you do all that properly
> what you've asked about, your study may qualify for publication in a
> pretty highly ranked journal.
> OTOH, maybe you're not planning to do it properly and/or you're
> underestimating the time needed for such modeling :-)
> To answer your question: I'm not aware of such possibilities in gretl
> (but as I said before I'm not the panel specialist here, and I have been
> corrected before).
> Gretl-users mailing list
One more question: my dependent variable arguably should be categorical. Is there any way to deal with this with gretl's panel routines, and if not, any reasonably doable way I could do this outside of the canned routines?
OK, I've looked over the command help for "arbond", and I think I
understand the basics, and this should do the trick. I do have a
couple of questions:
The help suggests that most indepedent variables should be
differenced, presumably because the D.V. is differenced by the A-B
method. It says that though in certain cases, such as time dummies,
you night want want differenced variables. What would you recommend
doing with other sorts of dummies? Two I have in mind are dummies to
represent intermittent exogenous events, specifically elections and
natural disaster (droughts and floods). I assume they should be differenced?
Also, this paragraph has me a little confused:
"By default the results of 1-step estimation are reported (with
robust standard errors). You may select 2-step estimation as an
option. In both cases tests for autocorrelation of orders 1 and 2 are
provided, as well as the Sargan overidentification test and a Wald
test for the joint significance of the regressors. Note that in this
differenced model first-order autocorrelation is not a threat to the
validity of the model, but second-order autocorrelation violates the
maintained statistical assumptions."
Which statistical assumptions does second-order autocorrelation
violate? Those of either procedure, or just the two-steps? And what
to do if estimation reveals significant second-order or higher
autocorrelations? By the way, is there any reason I'd want to use
1-step rather than 2-step?