Arfima modelling 1999

[Marius Ooms <ooms@econometriclinks.com>]

                                 Oxford/Rotterdam 31-5-1999
Dear professor,

You have expressed interest in modern long memory modelling.

Jurgen Doornik and Marius Ooms announce the launch of the new
Arfima package 1.00 for Ox 2.10. Ox 2.10 is a fast object-oriented
matrix language with a comprehensive mathematical and statistical
function library. The console version is free for downloading and 
works under many operating systems.

Theoretical details, empirical applications and numerous 
references are provided in two recent papers, also free for 
downloading. Ooms and Doornik (1999) discuss inference and 
forecasting for UK and US inflation. Doornik and Ooms (1999) 
document the procedures and provide computational details.

The Arfima package can now also be used interactively in a
point-and-click interface in connection with OxPack for Ox 
Professional 2.10 and GiveWin 1.20. This requires an MS
Windows Operating system (95/98/NT/...).

The Arfima package allows for fast estimation, simulation
and forecasting of stationary and nonstationary arfima models 
using likelihood based methods: exact ML, approximate conditional 
ML and modified profile likelihood. The mean of the process can be 
specified as a (non)linear function of explanatory variables.
Both optimal and naive forecasting methods are available. 
Restrictions can easily be tested and imposed.

The Arfima package comes with 13 sample programs to illustrate
specification, diagnostic testing, forecasting and simulation.
The package also contains data sets for monthly US and quarterly 
UK consumer price inflation. A long political data set based on 
weekly surveys on left-right orientation, accompanying Eisinga, Franses 
and Ooms (1999), International Journal of Forecasting, 185-199, is
used to show how to deal with (i.c. estimate) missing observations.

The Arfima package 1.00 and corresponding papers are now available 
for downloading from http://www.doornik.com. 
This link also provides recent information on Ox, OxPack and
GiveWin.

We thank all users for comments and suggestions regarding previous
versions of the package. Comments on the new version are welcome.

We are happy to demonstrate and discuss the possibilities of the
program in person when you visit the Society for Computational 
Economics Conference in Boston in June, the International Symposium on
Forecasting in Washington DC in June or the European Meeting
of the Econometric Society in Santiago de Compostela in August.
Please mailto:ooms@econometriclinks.com. if you are interested. Conference 
details are available through http://econometriclinks.com .

Here is a summary of the changes for users of previous versions 
  of the Arfima package:
- now using Durbin's algorithm for likelihood evaluation;
- improved starting value procedures;
- removed singularity for a single AR root at zero;
- using adjusted version of Durbin's algorithm for data generation;
- optimal forecasting for arbitrary d > 1;
- `additive' X, `innovative' Z-, and `weighting' W variables;
- automatic removal of non-invertable MA's;
- facility to simulate initial estimates;
- concentrating out regressors in Exact ML;
- addition of modified profile likelihood estimation for effective
  ``higher order asymptotic theory based'' bias correction for 
  processes with unknown mean.
- ArfimaSim class for Monte Carlo and Bootstrap analysis.
- general functions for the mean, possibly nonlinear in parameters;
- functions for communication with OxPack for GiveWin.

If you do not like to receive more messages on our work
on long memory modelling, please mailto:ooms@econometriclinks.com.

oxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxoxox
Dr Jurgen A Doornik, Nuffield College, Oxford
http://www.doornik.com
Dr Marius Ooms, Econometric Institute, Rotterdam
Editor Econometric Links The Econometrics Journal
http://www.econometriclinks.com
EctJEctJEctJEctJEctJEctJEctJEctJEctJEctJEctJEctJ