QUESTION:
What is an ARIMA model? Please explain in simple terms. No equations or math please.
ANSWER:
In
computing moving averages one needs to be concerned about two items:
1.
The number of periods to be used ( i.e. the length of the weights or the number
of weights )
and
2.
The actual values of the weights. The answer to this double-edged question is
called univariate
Box-Jenkins
which if correctly implemented returns both the number of weights to be used and
the actual
coefficients
to be applied to each lag. Modern approaches perform this task in a robust
manner such that
anomalies
in the data be they
a.
pulses b. level shifts c. seasonal pulses or d. time trends
are
accounted for thus providing a "good model".
A
productivity aid is available for this task is called AUTOBOX which rigorously
examines the possible
structures.
This software even enables the in-line testing of constancy of variance and
constancy of
parameters.
Univariate Box-Jenkins is also known as ARIMA and lots of material can be found
at this
website
and other places in cyberspace. There is no need to ASSUME the number of weights
and to
ASSUME
that all of the weights are equal.