Re: IMSL Statistics library


econstatsguy wrote:
Thanks a lot for the responses, Art and Homer! The auto_arima function
(in the C library, probably other languages too) is a powerful program
which, to quote
http://www.vni.com/solutions/forecasting/autoArima.html,
"...automatically:

* Estimates missing values within the data
* Incorporates the effects of outliers
* Performs seasonality adjustments
* Selects the best input parameters to an Autoregressive Integrated
Moving Average model [ARIMA(p,d,q)]
* Forecasts future values
* ARIMA provides:
* Automatic model selection from 6 alternative models
* Automatic missing value method selection from 4 alternative
methods
* Automatic seasonal adjustments
* Automatic outlier classification "

Isn't that really cool? :) Of course, once you get into the details of
an actual implementation with real data, problems/questions start
coming up, and I am hoping someone here is able and willing to offers
some tips.

I am currently involved in developing a program to monitor and optimise
an industrial process. Part of the problem is to obtain a reliable
algorithm to process data on machine performance. For now, I am using
a single set of measurements on one variable. The data are observed at
uneven intervals, which is one problem, but the IMSL routine can :fill
in" the missing data using a choice of methods. It then goes on to
estimate the parameters of a *seasonal * ARIMA model, essentially by
searching out the combination of parameter values which minimizes
Akaike's Information Criterion.

I already have the full documenation, but have not yet seen the program
code itself, so I am going by the description of the auto_arima
function in pages 555 et sequa of the IMSL C Statisitics Manual (URL
http://www.vni.com/books/dod/pdf/Cstat.pdf) .

My current question is a very simple one: how can I constrain the
method to *not* include seasonal factors? (At this point, seasonal
effects don't seem to make sense for this data, which is at roughly 5
minute intervals every day for 1.5 years ). From the optional
arguments listed inthe function description, it is not clear that this
can be done at all. The library has separate routines for Seasonal and
Non-seasonal ARIMA models, but those are *not* "automatic".

Any takers?

Homer, thanks for the pointer to the VNI forum- I will post this there
too.

Regards


Best of 6 Models ..that's a hoot ..

see http://www.autobox.com/pdfs/catchword.pdf for a discussion of that
topic ...

Software like AUTOBOX/FreeFore can be contrained by the user to do
precisely what you want. If you would like to chat with one of the
developers ....give me a call and I will try to help.

* Automatic outlier classification " Does it detect LEVEL SHIFTS ,
TREND CHANGES , THE ONSET OF SEASONAL PULSES or just PULSES .



Dave Reilly
Automatic Forecasting Systems
http://www.autobox.com
215-675-0652

.