Re: [Gretl-devel] Presentation of model results
El Sunday 23 November 2008 19:57:13 Riccardo (Jack) Lucchetti escribió:
On Sun, 23 Nov 2008, Sven Schreiber wrote:
> Am 23.11.2008 17:15, Allin Cottrell schrieb:
>> Mean dependent var 4.538837 S.D. dependent var 0.243346
>> Sum squared resid 1.617235 S.E. of regression 0.189575
>> R-squared 0.430985 Adjusted R-squared 0.405696
>> F(2, 45) 17.04203 P-value(F) 3.09e-06
>> Log-likelihood 13.26255 Akaike criterion -20.52509
>> Schwarz criterion -14.91149 Hannan-Quinn -18.40371
>> rho 0.018627 Durbin-Watson 1.960299
>>
>> (with the last row omitted for non-time series).
>
> Why actually have both the rho and the DW stat? (Assuming rho is the
> first-order autocorrelation coeff, right?)
I also don't see the necessity of rho (well, only for having an even number of
items here). As a former RATS user I miss, in the context of ARIMA models
(not for ols), the Ljung-Box statistic.
What I usually teach my students is that DW has just historical importance
and that if you want to test for autocorrelation the LM (Godfrey) test is
a much better tool. So, in principle I agree with you 100%. The only cases
(that I can think of right now) when I may want to actually see the DW
statistic are:
1) as always, for comparison/replication purposes
2) in the context of teaching what a spurious regression is (Granger &
Newbold's famous "rule of thumb")
3) in the context of cointegration tests, but IMHO nowadays Sargan and
Bhargava's idea is quite outdated too.
Hence, I wouldn't miss DW very much personally.
I vote for DW as well.
> Next, an unrelated suggestion: What do you think about grouping
the
> adjusted R-squared and the info criteria together, since conceptually
> they are more or less the same thing.
I prefer leaving the things as above. I consider the F() test may be
interpreted as a type of "significance test" for the R^2. Remember that
F=[R^2/(k-1)]/[(1-R^2)/(T-k)]
so I think it is natural that the F test goes inmediately after the R^2.
Agreed. Once we're at it, I'd like to express my preference
for the "AIC",
"BIC" and "HQC" acronyms, which shouldn't necessarily be
translated.
I also vote for AIC, BIC and HQC.
--
Ignacio Diaz-Emparanza
DEPARTAMENTO DE ECONOMÍA APLICADA III (ECONOMETRÍA Y ESTADÍSTICA)
UPV/EHU
Avda. Lehendakari Aguirre, 83 | 48015 BILBAO
T.: +34 946013732 | F.: +34 946013754
www.et.bs.ehu.es