Re: [Gretl-users] ADF result
by Sven Schreiber
[moving this over to the devel list]
Allin Cottrell schrieb:
>
> If I'm reading you right, you know that for gretl the answer is A
> (as indeed it is) but you doubt whether this is stated clearly
> enough.
>
Yes, that's exactly what I mean.
> My feeling is that it's clear in the documentation, but if can you
> suggest an economical way of cementing the correct interpretation
> in the actual test output (and bearing in mind that some people
> will run tests with lag order 24, which is why we currently write
> "..." for the lags!), please do and I'll be happy to consider it
> for inclusion.
>
Ok, for comparison here's the current output of an ADF test:
Erweiterte Dickey-Fuller-Tests, Ordnung 1, für PAU
Stichprobengröße 75
Nullhypothese Einheitswurzel: a = 1
Test mit Konstante
Modell: (1 - L)y = b0 + (a-1)*y(-1) + ... + e
Autokorrelationskoeff. 1. Ordnung für e: -0.015
geschätzter Wert für (a - 1): 0.0113244
Teststatistik: tau_c(1) = 3.82463
asymptotischer p-Wert 1
How about changing the model ("Modell") line just a little bit:
Modell:
(1-L)y = b + (a-1)*y(-1) + ... + a_[lag order]*(1-L)y(-[lag order]) + e
That way it's absolutely clear that the order refers to the lagged
differences. BTW, I removed the blanks from (1 - L) and changed b0 to b
on purpose, but of course that's just cosmetic.
Talking about cosmetics: maybe the recent reformatting of a model's
output should also be done here? I mean something like:
Test mit Konstante
Modell: (1-L)y = b + (a-1)*y(-1) +...+ a_1*(1-L)y(-1) + e
Autokorr'koeff. 1. Ordnung für e: -0.015
geschätzter Wert für (a-1): 0.0113244
Teststatistik: tau_c(1) = 3.82463
asymptotischer p-Wert: 1
Of course all this is just a first shot, not perfectly thought through.
thanks,
sven
16 years
garch dialog
by Sven Schreiber
Today I've noticed (teaching again...) that in the GARCH dialog window
it says "ARCH p: .... ARCH q:". I guess the latter should rather be
"GARCH q", no?
-sven
16 years
questions on Durbin-Watson stuff
by Allin Cottrell
I think I'm finished with Durbin-Watson for the moment: there's
now a --dw-pval option to OLS and fixed effects estimation, which
tells gretl to compute the p-value for the D-W statistic (using
the Imhof method), hence making this statistic more useful (albeit
at the cost of computations which could be quite time-consuming
for big datasets).
In the course of developing this I added 3 functions (and
documented 2 of them). But I wonder if it's worth keeping any of
them and I'd like to hear people's preferences. (To my mind, we
don't want to multiply functions unnecessarily, since this makes
it harder to find what you're looking for.) The functions in
question are:
* dwmat (takes one scalar argument, n): produces a tridiagonal
Durbin-Watson matrix of order n. This can be done with standard
matrix functions but it's a bit fiddly. Would anyone like to see
this kept as a public function in "genr"?
* dwpval (two matrix arguments, u and X): computes the p-value for
the Durbin watson statistic given a column vector of residuals, u,
and a regressor matrix, X. Probably too specialized to keep, if
we now have the --dw-pval option?
* imhof (matrix m, scalar x): computes Prob(u'Au < x) for a
quadratic form in standard normal variates, u. If the argument m
is a square matrix it is taken to be 'A', otherwise if it's a
column vector it is taken to be the precomputed eigenvalues of A
(otherwise error). This is not documented at present, but it's
more general than the other 2 functions and maybe is worth
documenting and keeping?
Allin
16 years
Durbin-Watson p-values
by Allin Cottrell
One new thing for testing. I know this is not to everyone's
taste, but I've added a function for computing the p-value for the
Durbin-Watson statistic (using the Imhof method). The idea in the
background was to get a correct p-value for D-W in the panel
context, though that is not fully implemeted yet.
You can try it out thus (a simple time-series, non-panel
illustration):
<script>
open theil.gdt
ols consume 0 relprice
matrix u = $uhat
matrix X = { $xlist }
scalar pv = dwpval(u, X)
ols consume 0 income relprice
matrix u = $uhat
matrix X = { $xlist }
scalar pv = dwpval(u, X)
Compare http://shazam.econ.ubc.ca/intro/testauto.htm#auto1
Allin
16 years
Placeholder string
by Cristian Rigamonti
Hi, maybe you have forgotten to replace the string
"note on model statistics abbreviations here"
in lib/src/modelprint.c:2266 with the intended string.
(I've found it in the latest .po file)
Cri
--
GPG/PGP Key-Id 0x943A5F0E - http://www.linux.it/~cri/cri.asc
16 years
