Dnia Fri, 29 Feb 2008 08:15:47 -0800 (PST)
cheesekeeper <cheesekeeper@[EMAIL PROTECTED]
> napisa=C5=82(a):
> Greetings,
>=20
> I will be forecasting data that ****fts direction at points due to
> changes in business decisions. I would like to get an idea of the best
> forecasting model to use for extrapolating future results.
>=20
> Basically, numbers may have gone up for 6 months are a relatively
> steady rate, and then due to a ****ft in the business, they went down
> at a relatively steady rate for 9 months. I would like to use a
> forecasting model that will essentially "discard" the 6 months of
> increases, and consider what appears to be the current trend of 9
> months of decreases.
>=20
> I am currently using a weighted moving average, giving weight to more
> recent numbers, but that works better for data with changes in trends
> further in the past. If instead of 9 months of steady decreases, there
> were only 3 months of steady decreases, the model would not be as
> good, since the period of increases would have a greater effect.
>=20
> Is there a method that will identify those ****fts in direction, and
> ignore or minimize the effect of previous trends?
>=20
Box-Jenkins methodology with intervention analysis and
outliers detection should help you.
Best regards
--=20
[ Wit Jakuczun <W.Jakuczun [at] wlogsolutions.com> ]
[ WLOG Solutions http://www.wlogsolutions.com
]
|
