[S] QUESTION: ACF and fr-ARIMA-models

Oliver Klinger (klinger@lkn.e-technik.tu-muenchen.de)
Mon, 14 Sep 1998 11:59:28 +0200


Hello, Dear S+ers!

THE SITUATION:
I'm trying to model a long-memory-timeseries using the arima.fracdiff
and arima.fracdiff.sim-functions of S+ 4.5
I decided to use models of order (2, d, 2) (i.e. 2 AR-parameters and 2
MA-parameters)
The timeseries to be modeled is a measurement of LAN bandwidth used
during a videoconference (LiveLan).

THE PROBLEM:
When simulating a new timeseries with arima.fracdiff.sim and as model
the model calculated with arima.fracdiff, then the simluated timeseries
shows an extremely bad short-range-dependence. This can be seen in the
autocorrelation function (ACF) of the simulated series. In the acf-plot
there are at small lags (<50) extrremely low correlations to be seen
(~0.01 whereas they should be ~0.1 ... 0.25!)

THE QUESTION:
What can/must I do to improve the short-range behaviour of the simulated
timeseries without changing the order of my fr-ARIMA-models?!?!?

*URGENT*

Thanx in advance

Oliver Klinger
Institute for Communication Networks
Technical University of Munich

-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news