|Date: ||Wed, 17 Nov 2004 15:40:52 -0800|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|From: ||"David L. Cassell" <cassell.david@EPAMAIL.EPA.GOV>|
|Subject: ||Re: pulsar algorithm?|
|Content-type: ||text/plain; charset=US-ASCII|
"F. J. Kelley" <jkelley@UGA.EDU> wrote:
> I have been asked about a program or algorithm called Pulsar
> and whether this is available (or someone has implemented it)
> in SAS.
The things called Pulsar algorithms typically look like this:
 fit a curve to time series data;
 standardize the residuals by dividing by the standard error;
 plot and see if anything 'looks like a pulse'.
If my description seems less than enthusiastic, it is because
*I* am less than enthusiastic about Pulsar algorithms. You can do
this yourself in SAS (pretty much every regression-like proc
lets you get the Y, the predicted Y, and the standard error so
you can 'standardize' and plot it out). Or you can search
Google for "Pulsar algorithm" and find stuff there.
But the idea is crucially dependent on fitting the right model!
If you have time series data and don't fit the right model, your
residuals need be independent, much less have the right standard
errors. If you have something like a GARCH model, then assuming,
say, a simple ARMA(1,1) model will muck things up because you
aren't dealing with the time-series structure of the variance.
This will give you the appearance of a big pulse in your Pulsar
algorithm, although it may just be more fluctuation in your Y.
So drop this on your end-users and let them make the call.
David Cassell, CSC
Senior computing specialist