Is it possible to simultaneously estimate mean and GARCH parameters in R? In other words, I would like to estimate the normal regression equation Y = b X + u and simultaneously do a garch process on the u's to correct the standard errors. I was thinking maybe something with systemfit(), but I can't quite come up with it. Thanks, Tobias --
Dirk Eddelbuettel
2005-May-19 23:14 UTC
[R] Simultaneous estimation of mean and garch eq'n
Tobias Muhlhofer <t.muhlhofer <at> lse.ac.uk> writes:> Is it possible to simultaneously estimate mean and GARCH parameters in R? > > In other words, I would like to estimate the normal regression equation > > Y = b X + u > > and simultaneously do a garch process on the u's to correct the standard > errors. > > I was thinking maybe something with systemfit(), but I can't quite come > up with it.It is but you have to write out the loglikelihood (and possibly its gradient) which you could then maximise with optim() and friends. Adrian's GARCH(1,1) is hard-coded with an analytic gradient -- the convenience of having this powerful routine pre-made and comes at the price of its lack of flexibility. Diethelm's fSeries from Rmetrics can estimate Garch models by calling Ox. That may work for you too. Hope this helps, Dirk
This might be one of those situations in which you should say what what you are trying to do rather than how you are trying to do it. It is my understanding that estimates of b are asymptotically well behaved in this situation, at least for b<1. If, however, you are trying to get CI's for b in finite samples conditioned upon the errors being generated by a GARCH process then this is a different issue. HTH Phineas>>> Tobias Muhlhofer <t.muhlhofer at lse.ac.uk> 05/19/05 9:56 PM >>>Is it possible to simultaneously estimate mean and GARCH parameters in R? In other words, I would like to estimate the normal regression equation Y = b X + u and simultaneously do a garch process on the u's to correct the standard errors. I was thinking maybe something with systemfit(), but I can't quite come up with it. Thanks, Tobias -- ______________________________________________ R-help at stat.math.ethz.ch mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
It is my experience that location parameters are not very affected by the garch parameters. So doing a naive estimate of location, followed by the garch estimate, followed by an estimate of location accounting for heteroskedasticity is likely to be indistinguishable from the estimates from the full likelihood. If you compare garch with the naive estimate of location versus the garch estimate with no estimate of location, you are probably not going to see much difference. That difference is likely to be smaller than if you change the time period of estimation slightly. Patrick Burns Burns Statistics patrick at burns-stat.com +44 (0)20 8525 0696 http://www.burns-stat.com (home of S Poetry and "A Guide for the Unwilling S User") Tobias Muhlhofer wrote:> Is it possible to simultaneously estimate mean and GARCH parameters in R? > > In other words, I would like to estimate the normal regression equation > > Y = b X + u > > and simultaneously do a garch process on the u's to correct the > standard errors. > > I was thinking maybe something with systemfit(), but I can't quite > come up with it. > > Thanks, > Tobias > -- > > ______________________________________________ > R-help at stat.math.ethz.ch mailing list > https://stat.ethz.ch/mailman/listinfo/r-help > PLEASE do read the posting guide! > http://www.R-project.org/posting-guide.html > > >