Re: Adding Regressors to ARIMA.

On Jun 20, 12:45 pm, aruzinsky <aruzin...@xxxxxxxxxxxxxxxxxxxx> wrote:
On Jun 20, 5:44 am, d...@xxxxxxxxxxx wrote:





On Jun 19, 1:16 pm, aruzinsky <aruzin...@xxxxxxxxxxxxxxxxxxxx> wrote:

On Jun 19, 8:15 am, Paul Rubin <ru...@xxxxxxx> wrote:

aruzinsky wrote:

SNIP ...

As far as I know, Box and Jenkins never compared the ARIMA model to
conventional regression when  in reality, Yt = a*t + b + Et, where Et
is iid noise, t is time, and a and b are unknown.  In this case, ARIMA
differencing produces an MA process with zeros on the unit circle..  As
I recall, Box and Jenkins swept under the rug all cases with transfer
function zeros on the unit circle.

The earlier work by Box-Jenkins did avoid this issue. Later
researchers incorporated determinstic variables such as "t" or level
shifts or seasonal pulses or pulses into the Transfer Function. For
details on this you might review the W.S. WEI book on time series
(Chapter 9 ) ISBN 0-201-15911-2 Addison-Wesley.

We at AFS have been identifying and incorprating such structure for
over 20 years using INTERVENTION DETECTION procedures to add to the
normal stochastic structure of the TF.

for more on this you might go tohttp://www.autobox.com/andgoogle
the site . You might also enroll athttp://www.autobox.com/AFSUniversity/afsuFrameset.htm
and review some of the URL's under Intro to Forecasting

dave reilly
AFShttp://www.autobox.com

If you read my first post in this thread, you will see that I claimed
to invent or reinvent incorporating functions of t circa 1988 into the
AR model which predates the W.S. WEI book.

You at AFS still haven't answered the most obvious question, "Do you
know how to estimate MA parameters with zeros on the unit circle?"
Come on, "yes" or "no?"- Hide quoted text -

- Show quoted text -

If your findings are correct , I suggest that you submit your
"invention" to a refereed journal so we can move the science
forward ..

Dave



.

Relevant Pages