QUESTION:
Why do I need Box-Jenkins? Can't I use exponential smoothing or a simple average of the past, like a 12
period moving average?
ANSWER:
All of the approaches that you mentioned are forms of Box-Jenkins. An ARIMA model, univariate Box-Jenkins, is simply the answer to a forecaster's dream;
How much of the series do I need to use
and
What weights should I use?
An exponential smoothing model assumes that all of the past is needed and that the weights can be
described by one curve or equation. This is not a very realistic assumption and each and every time series
may not be adequately described by these two assumptions. A simple 12 period average assumes that 12
previous values are needed and that these weights are equal to each other (.083). A possible scheme might
be 2/6, 1/6, 3/6 or whatever. There is no need to assume the optimal way to weight the past let analysis i.e.
Box-Jenkins do the work and use the best conditions.