|
PureBytes Links
Trading Reference Links
|
Bob,
the problems you indicate are real, and have been the reason for me to
totally abandon TS in 1999 in favor of Mathematica. I now do all of my
testing and developing in this extremely versatile language, with no
inclination of ever returning to EL again.
Here is what I posted on this list on April 12, and you may see from this
that I am in full agreement with your analysis:
"As far as Mathematica is concerned ..., the language actually has
*arbitrary* floating point precision built-in, enabling you to specify the
desired internal accuracy of a calculation. This is important because of the
issue of floating-point error propagation. As you know this problem can lead
to totally meaningless results which may be wrong in all places of the
mantissa, and sometimes can even be several orders of magnitude away from
reality.
"It always surprises me when I see people use Metastock or Tradestation for
calculation-intensive number crunching while being entirely at the mercy of
the built-in accuracy, or lack of it, of these packages.
"We had such a case some time ago on the Metastock list where one poster got
into big arguments with Metastock support because the software would give
him signals different from those he was getting with Excel. The reason
turned out to be some decimal in a calculation which finally lead to a
different result, caused by different floating-point implementations [single
vs. double precision]. And these calculations were just simple Moving
Averages, not Fourier analysis or the inverting of Hilbert matrices!
"So this issue of accuracy is not to be taken lightly, IMHO."
Best wishes,
Michael Suesserott
|