Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: RE: RE: RE: Not the same results with GARCH


From   Nick Cox <njcoxstata@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: RE: RE: Not the same results with GARCH
Date   Wed, 26 Oct 2011 12:27:52 +0100

As said, we don't have your data. But it seems that there are two
major peaks to be found in the surface being explored. I guess you
should worry a bit about how weird the topology (topography) of the
surface being explored really is, not just about what the algorithms
are finding but about what they are not showing you.

These models are very attractive in many ways. My gut feeling is that
they have been a bit oversold and in any case it is not a great
surprise that they may be difficult to fit. The history of time series
analysis is full of models that went in and out of favour almost as
rapidly as zoot suits or hula-hoops.

Nick

On Wed, Oct 26, 2011 at 11:02 AM, Valerie Orozco
<Valerie.Orozco@toulouse.inra.fr> wrote:
> Thank you Nick for your answer.
>
> You're true, I have to consider some criteria to find my best model. But I still find really surprising that the 4 standard optimization techniques give the same results and that the default one (combination of 2) gives very very very strange and different results...
>
> Fortunately the "best" models (looking at the LL, BIC and AIC criteria, see below) are those that seem the most credible in my opinion.
>
> The conclusion to this is that it is really important to estimate the model with all the optimization techniques to check robustness and choose the best one.
>
> --------------------------------------------------------------------------------------------
>                         Free                    nr                  bhhh                dfp                   bfgs
> --------------------------------------------------------------------------------------------
> valeur
> _cons               110.1***        102.3***        102.3***        102.3***        102.3***
> --------------------------------------------------------------------------------------------
> ARMA
> L.ma                0.781***        0.695***        0.695***        0.695***        0.695***
> --------------------------------------------------------------------------------------------
> ARCH
> L.arch            -0.0279*          0.241**         0.241***        0.241**         0.241**
> L.garch            -0.956***        0.742***        0.742***        0.742***        0.742***
> _cons               254.3***        5.671           5.676*          5.672           5.678
> --------------------------------------------------------------------------------------------
> aic                1979.8          1972.9          1972.9          1972.9          1972.9
> bic                1997.5          1990.7          1990.7          1990.7          1990.7
> LL                 -984.9          -981.5          -981.5          -981.5          -981.5
> --------------------------------------------------------------------------------------------
>
> -------------------------------
> Valérie OROZCO
> Toulouse School of Economics (INRA-GREMAQ)
> 21, allée de Brienne
> F-31000 Toulouse, France
>
> MF 219
> +33 5 61 12 85 91
> -------------------------------
>
> -----Message d'origine-----
> From      Nick Cox <n.j.cox@durham.ac.uk>
> To        "'statalist@hsphsun2.harvard.edu'" <statalist@hsphsun2.harvard.edu>
> Subject           st: RE: RE: Not the same results with GARCH
> Date      Wed, 26 Oct 2011 10:24:51 +0100
>
> No one replied yet, no doubt partly because we can't say anything about your data.
>
> My guess is that you did nothing wrong; these models are not always as successful as people want, and different search methods can find different (sub-)optima within parameter space.
>
> You don't cite any figures of merit/badness-of-fit/goodness-of-fit measures, but there seems to be a choice between two quite different fits. Perhaps there are also graphical or scientific grounds to choose between them.
>
> Nick
> n.j.cox@durham.ac.uk
>
> -----Message d'origine-----
> De : Valerie Orozco
> Envoyé : mercredi 26 octobre 2011 08:39
> À : 'statalist@hsphsun2.harvard.edu'
> Objet : RE: Not the same results with GARCH
>
> I just noticed that the optimization technique used in my "Free" model (without any technique specified) is :
> e(technique) : "bhhh bfgs"
> That is a combination of 2 techniques. I don't understand why the results are so different from the others estimations but it explains why the results are not the same than the "nr" ones.
>
> valérie
>
> -------------------------------
> Valérie OROZCO
> Toulouse School of Economics (INRA-GREMAQ) 21, allée de Brienne
> F-31000 Toulouse, France
>
> MF 219
> +33 5 61 12 85 91
> -------------------------------
>
> -----Message d'origine-----
> De : Valerie Orozco
> Envoyé : mardi 25 octobre 2011 15:38
> À : statalist@hsphsun2.harvard.edu
> Objet : Not the same results with GARCH
>
> Hi,
>
> I have some trouble with the "arch" command.
> I want to model a MA(1) process and a GARCH on its residuals. When I specify the technique of optimization, I find quite the same results whatever the method (nr bhhh dfp bfgs). But when I don't write the "technique" option, I have very strange results (see below).
> I thought that no specifying the technique means that the technique would have been the default one (nr) but it seems not...
> Do you have an idea? What I am doing wrong???
>
> Thank you.
> Valérie
>
> Here you have my program:
> /****************************************************/
> eststo clear
> eststo Free : arch  valeur, ma(1) garch(1) arch(1)
>
> local tech "nr bhhh dfp bfgs"
> foreach t of local tech {
>    qui eststo `t' : arch  valeur, ma(1) garch(1) arch(1)  technique(`t') }
>
> estout
> /****************************************************/
>
> And here, you have the results :
> -----------------------------------------------------------------------------
>                     Free           nr         bhhh          dfp         bfgs
>                        b            b            b            b            b
> -----------------------------------------------------------------------------
> valeur
> _cons            110.1188     102.2961     102.2969     102.2965     102.2959
> -----------------------------------------------------------------------------
> ARMA
> L.ma             .7813551     .6950596     .6950131     .6950479     .6950538
> -----------------------------------------------------------------------------
> ARCH
> L.arch          -.0278845     .2406442     .2407423     .2405665     .2407807
> L.garch         -.9560793      .741724     .7416003     .7417785     .7415708
> _cons            254.3421     5.670839     5.675977     5.672461      5.67767
> -----------------------------------------------------------------------------
> -------------------------------
> Valérie OROZCO
> Toulouse School of Economics (INRA-GREMAQ) 21, allée de Brienne
> F-31000 Toulouse, France
>
> MF 219
> +33 5 61 12 85 91
> -------------------------------
>
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index