Email follow ups of the fixed d problem noted that under Windows
this converges, but not under Unix.
The different behaviour between Unix and Windows is due to
missing values (NaN). The Windows version, when a NaN is returned
as a function value, MaxBFGS will enter the line search until a
valid function value is obtained. Under Unix this doesn't work
because of slightly different treatment of NaNs.
Modified maximization code which checks for missing values
is available from
http://www.nuff.ox.ac.uk/Users/Doornik/download.html#oxfixes.
oxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxox
Dr Jurgen A Doornik
Nuffield College
Oxford OX1 1NF, UK
tel. +44-1865-278610 fax +44-1865-278557
http://www.nuff.ox.ac.uk/Users/Doornik/
http://www.oxedit.com
http://www.pcgive.com
oxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxox
[log in to unmask] wrote:
> I've used version 0.77 of the arfima package to estimate both
> arfima and arma models and now have upgraded to 1.00 of the
> arfima package. I encounter no problems when estimating arfima
> models but when I try to use FixD(0) to estimate a arma model
> the results from Estimate() show that the initial starting value
> of d equals zero, but the parameter estimates include a non-zero
> estimate of the differencing parameter. There is something clearly
> wrong with this, is anyone else aware of this problem or how to
> fix it. I haven't yet taken a look at arfima.ox to determine if it
> is a coding mistake, I'm hoping to avoid that with this message
> thanks.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|