QUESTION:
Alright, so your
computation of the standard deviation is more correct, explain why I need
AUTOBOX to
detect level shifts,
seasonal pulses and time trends.
ANSWER:
If one restricts OUTLIER
DETECTION to detecting unusual values that are INDEPENDENTLY flagged,
one can miss important information. One man's
signal is another man's noise is the best way that I can
put it. If you have an "UNUSUAL"
value every December, this is "USUAL" since it represents a
predictable process. Software that
INDEPENDENTLY evaluates each and every value based only upon IT's
characteristics can severely miss the boat in
picking up quality information and predictable structure.
This is then the argument for Seasonal Pulse
identification, a unique feature of AUTOBOX. The idea
here is to test for instantaneous and
immediately disappearing changes in the mean of the errors. If the
mean of the errors changes to another
"regime" this is called a level or a step shift. For example a series
1,1,1,1,1,2,2,2,2,2,2,2, would exhibit A STEP
SHIFT or a change in the mean. AUTOBOX would identify
the need for a dummy variable that had 0,0,0,0,0,1,1,1,1,1,1,1
and when estimated the coefficient of
response would be a 1 with an intercept of 1.
A step or level shift can be seen to be the integration of a
pulse, 1 1 0 1 0 1 0 1 0 2 when differenced 1
2 0 2 0 2 0 2 0 2 0 2 0. In the same spirit, a trend can be seen
to be the integration of a step 1 1 2 1 3 1 4
1 5 1 7 when differenced 2 9 2 11 2 13 2 15 2 17 2 19.
AUTOBOX can then detect the need for a TIME
TREND VARIABLE which of course is nothing more
than the integration of a step or the double
integration of a pulse.