PureBytes Links
Trading Reference Links
|
The problem with single output optimization is that it can create unknown
peaks in your optimization. For example, say you optimize a simple moving
average system against Total Profit for bonds. The best value is 15 days.
But then looking at the actual rows of optimization data output for Total
Profits, you might see that 15 worked great but 16,17,18, 14,13,12 worked
horribly! So next quarter when you used that spike of 15 you would be, in a
word, hosed. The best thing to do is look at ALL the rows of data for each
optimization output and pick a high plateu for each and check the cross
correspondence between them all. Also known as cluster analysis.
Complicated? Yep. Necessary? Absolutely.
Fortunately there is a nice product out there by Rina Systems (and no, I
don't work for Rina nor do I get any kind of compensation for this :) that
produces 3-d graphs of optimization reports across all outputs. You can
then do your cluster analysis from there visually. It is a very very cool
and useful tool. It goes for about $500. A real steal because even if you
pay yourself $25 an hour for doing cluster analysis manually pouring over
rows of data, after looking at optimization reports for even 5 systems you
will have broken even. Not to mention the money spent on your increased
lense perscription after looking at those hundreds of rows of data over and
over and over...
Whew. Thank goodness system development is easy! Time to move onto my next
billion trillion dollars systems :)
ianwaugh@xxxxxxxxx
link.co.uk (Ian To: omega-list@xxxxxxxxxx
Waugh) cc: ianwaugh@xxxxxxxxxxxxxxxxxxx
Subject: System optimisation
02/19/2001 12:14
PM
Please respond to
ianwaugh
In TS' Tools>Options dialog there's a setting for Optimization where you
can optimise a system by most of the criteria you see in the System
Performance Summary report.
I've just read Thomas Stridsman's book, Trading Systems That Work, and he
suggests that few of the TS reports are of much use. I was wondering
what's the best option for optimising a system? Total net profit seems to
make the most sense but I've tried Avg trade (win+loss), % profitable, and
profit factor among others. They each produce slightly different results.
Is there a generally accepted "best" setting to use? If not, I'd
appreciate hearing members' views and preferences.
Ian
|