Thanks Sven, you did answer my question: I tried to reproduce what you describe "by hand" and I come up with identical values as Gretl's ADF procedure. I was stupid not to figure it out by myself: of course we cannot compare AIC based on different sets of data. Just for completeness I include the example you asked for. All the best, Matteo
PS: Here is the example
k = 4: AIC = -324.082
k = 3: AIC = -325.356
k = 2: AIC = -327.350
_ k = 1: AIC = -329.349__
_ k = 0: AIC = -322.373
Augmented Dickey-Fuller test for IBO
[...]
Augmented Dickey-Fuller regression
OLS, using observations 1974:3-1987:3 (T = 53)
Dependent variable: d_IBO
coefficient std. error t-ratio p-value
-------------------------------------------------------
const 0.00994959 0.00667146 1.491 0.1421
IBO_1 −0.0686761 0.0415984 −1.651 0.4564
d_IBO_1 0.389366 0.124332 3.132 0.0029 ***
_AIC: -345.238_ BIC: -339.327 HQC: -342.965
Il 16/06/2017 18:24, Sven Schreiber ha scritto:
> Am 16.06.2017 um 13:00 schrieb Matteo Pelagatti:
>> Dear Gretl users and developers,
>> I noticed that the AIC used in the lag selection procedure of the ADF
>> test is different from the one reported under the regression results.
>> Also, the lag selection would be different if based on the
>> regression's AIC.
>>
>> My questions are:
>> "What formula is used for the selection procedure?",
>> "Why does it differ from the usual one?"
>
> It might help if you could give a concrete example. Until then, my
> guess is that this is the same issue as in Ignacio's question just
> now: In the test-down procedure the (shortest) sample is kept fixed
> for all lags, while in the end result all available obs are used, thus
> leading to differences. This would not be gretl's fault.
>
> cheers,
> sven
> _______________________________________________
> Gretl-users mailing list
> Gretl-users(a)lists.wfu.edu
> http://lists.wfu.edu/mailman/listinfo/gretl-users
>
--
Matteo Pelagatti
Department of Economics, Management and Statistics
University of Milano-Bicocca
Via Bicocca degli Arcimboldi, 8
20126 Milano
Tel +39 02 6448.5834
http://www.statistica.unimib.it/utenti/p_matteo/
|
Thanks Sven,
you did answer my question: I tried to reproduce what you describe "by hand" and I come up with identical values as Gretl's ADF procedure. I was stupid not to figure it out by myself: of course we cannot compare AIC based on different sets of data. Just for completeness I include the example you asked for. All the best, Matteo PS: Here is the example k = 4: AIC = -324.082 k = 3: AIC = -325.356 k = 2: AIC = -327.350 k = 1: AIC = -329.349 k = 0: AIC = -322.373 Augmented Dickey-Fuller test for IBO [...] Augmented Dickey-Fuller regression OLS, using observations 1974:3-1987:3 (T = 53) Dependent variable: d_IBO coefficient std. error t-ratio p-value ------------------------------------------------------- const 0.00994959 0.00667146 1.491 0.1421 IBO_1 −0.0686761 0.0415984 −1.651 0.4564 d_IBO_1 0.389366 0.124332 3.132 0.0029 *** AIC: -345.238 BIC: -339.327 HQC: -342.965 Il 16/06/2017 18:24, Sven Schreiber ha scritto: Am 16.06.2017 um 13:00 schrieb Matteo Pelagatti:
-- Matteo Pelagatti Department of Economics, Management and Statistics University of Milano-Bicocca Via Bicocca degli Arcimboldi, 8 20126 Milano Tel +39 02 6448.5834 http://www.statistica.unimib.it/utenti/p_matteo/ |
