How widely spaced are these 375 x-values?

If the x-values are 1,2,3,... then fitting a 6th degree polynomial is a very
difficult problem numerically (condition number ~3E31). The direct algorithm
used by LINEST prior to 2003 would likely produce meaningless results. The
chart trendline and LINEST might be accurate to a few figures, but the
problem could be challenging for them as well. To reliably fit a problem
this numerically difficult, you might need a package that uses quadruple
precision (I don't know of any statistics programs that do) or arbitrary
precision (cf.
http://groups.google.com/group/micro...fdea49d5c999a7
). You could probably do this in double precision by fitting orthogonal
polynomials, assuming that the x-values themselves are really accurate enough
to be worth the effort.

Your mention of "the formula provided by the graph" raises another issue:
while the chart trendline (unchanged with 2003) has always been quite good
numerically (better than almost all dedicated statistical packages, except
where they fit by orthogonal polynomials), by default very few figures of
this high quality fit get displayed on the graph. You need to right click on
the equation and change the numeric format to display scientific notation
with 14 decimal places.

Jerry

"David" wrote:

Hi Jerry,

I usually have somewhere around 375 data points, which goes up by 1 each
week, as I am tracking a stock closing, but only on a weekly basis. Yes, it
is the 6th degree polynomial that I add to the graph. I tried to figure out
the individual data points represented by the graph, using the formula
provided by the graph, but the "point" was not very accurate and I am given
the understanding that this was due to rounding errors and the precion of
Excel, prior to Office 2003.

Maybe this has all changed? I had also run across a third party add-on that
was claiming to have increased the accuracy, but did not purchase it and have
not heard anyone else mention it or vouch for it. But if I could find a way
to figure out the individual points represented by the graph, I would be
interested in doing that.

Thanks for your help.


--
David


"Jerry W. Lewis" wrote:

If by "6th degree", you mean a 6th degree polynomial, then you should be
conserned about whether you have sampled a wide enough range of data to be
able to reliably estimate the parameters. Even with independent parameters
and/or adequate data range, validation of the formula could be interesting.
Remember the famous quote of von Neumann "With four parameters I can fit an
elephant, and with five I can make him wiggle his trunk."

Jerry

"David" wrote:

I just started reading your thread and have become interested. I was at one
time trying to create data points from an anaysis of Stock closiings, that
are represented by a regression (6th degree) line. I was not able to use the
formula supplied in previous versions of Excel, prior to Ver 2003. But I
would be interested in trying this again, since i have the new version. I
would like to actually create data points as respesented by the regression
line, fi that is possible?

Thanks,
--
David