|
PureBytes Links
Trading Reference Links
|
Pierre,
There aren't many tried and true algorithms for finding chart patterns that
are reliable. Nothing like "sorting" algorithms. Digitizing peaks and
troughs is easy enough but chart patterns with their infinite variation in
shape, size, and timing is what makes the whole thing difficult. NASA, JPL
or someone at NSA with access to a Cray could probably whip up something
that would work quite reliably. Unfortunately, most mere mortals don't have
that level of access, and we do the best we can with basic charting
software.
Most of the software that we use was conceived long before computer
detection and analysis of chart patterns was thought of. We are limited to
the primitive 'macro' sort of programming that has been added on to what
started out as just charting software. Previously, spotting patterns and
their relevance to the price situation at hand, was purely a human endeavor,
wrought with individual interpretation.
If I were going to develop something from the ground-up, I'd take a look at
speech recognition software algorithms. Spectra of speech is as varied as
price charts among dialects, but the software still has to come up with
"CAT" when it is spoken.
Another stab at the problem would to look at gas chromatography software. I
was a chemist and used Hewlett-Packard Chemstation software on a daily
basis. This software used various adjustable algorithms to draw baselines
from peak to peak. Then, another algorithm would integrate the peak to give
relative areas under each peak. It was amazingly accurate. One could
extrapolate the timing, placement, and size of these peaks as some sort of
pattern finder. On your next lunch break, cruise over to the Chem Dept. and
see what they are using.
Thanks,
Corey Saxe
csaxe@xxxxxxx
----- Original Message -----
From: "Pierre A. von Kaenel" <pierre@xxxxxxxxxxxxx>
To: <metastock-digest@xxxxxxxxxxxxxxxxxx>
Sent: Tuesday, November 27, 2001 11:27 AM
Subject: Pattern recognition using EOD quotes
> I've got a student who'll be doing an undergraduate senior thesis this
> spring and have been passing on ideas for thesis topics. One that crossed
> my mind recently is applying basic pattern recognition concepts to
> end-of-day quotes (open-hi-lo-end-volume) to find chart patterns such as
> trend lines, head & shoulders, double/triple tops and bottoms, etc. Would
> anyone know whether the algorithms used in some charting software to find
> some of these basic patterns were developed by trial-and-error methods or
> whether there are known theories or concepts used to devise algorithms?
I
> would prefer that the work be based on some known principles of pattern
> recognition rather than based on intuition. Of course, if there isn't any
> applicable theory to use as a foundation, perhaps the student can try to
> develop some basic principles.
>
> Thanks for any input.
>
>
> ***************************************************************
> Pierre A. von Kaenel
> Assoc. Prof., Math & Computer Science Dept.
> Skidmore College
> Saratoga Springs, NY 12866
>
> pierre@xxxxxxxxxxxxx
>
> "Simplify, simplify, simplify" - Thoreau
> ***************************************************************
>
|