[EMAIL PROTECTED] (Tugba Taskaya) wrote in message news:<[EMAIL PROTECTED]>...
> It says that a(t) should be independently distributed as normal random
> variables with mean=0. So can't I find an initial residual that makes
> the mean of the residuals 0?

This means that after you compute the a(14),a(15),......a(t)

you need to test the the ACF ( autocorrelation ) of these values has
no significant structure ....

and that the mean of the a's is zero everyhere ( i.e. for all
contiguous subsets ) otherwise you need to introduce dummy variables
to deaal with this Gaussian violation ( Pulse,Step, Seasonal Pulse
and/or Local Time Trends )

and that the variance of the a's is constant over time

and that the estimated parameter(s) (thetha 1 in your case) is
constant over time.

> 
> On the paper, they say that they aim to take first-order regular
> difference and the first seasonal difference in order to remove the
> growth trend and the seasonality characteristics.


So they say ! Observing non-stationarity is one thing ...developing
the correct remedy to make the series stationary is the difficult step
...

In the early days of Box-Jenkins development practictioners were
directed to
difference to deal with these symptoms. Modern-day approaches also
consider deterministic components such as Pulse,Step, Seasonal Pulse
and/or Local Time Trends as identified via INTERVENTION DETECTION .


I suggest that you search DOWNLOAD.COM or TUCOWS.COM for a FREEWARE
program called FreeFore which automatically tests for and remedies for
all the requirements underlying the model including all that I have
mentioned in this post.

regards

Dave R
Automatic Forecasting Systems
215-675-0652


P.S. If I can help privately please call or let me know when you can
call ...


> 
> thanks.
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to