|
PureBytes Links
Trading Reference Links
|
Maybe I made some mistakes along the way but I spent weeks working on
walk-forward optimization and testing results out of sample for several systems
and several markets. Looked elegant enough. It was a complete failure.
My conclusion was that if you have a self-adapting system then that adaptation
is better placed upon the building blocks of the system rather than on the
system results. Dangerous to use optimation other than to determine that
profitable system variables are within a reasonable range of expectation.
(Perhaps even minor tuning but rarely.) Of course, you would have to have a
logical apriori basis for forming that range of expecation! For example, some
number of standard deviations. If you expect the minority of a sample to occur
and your occurance is greater than .5 standard deviations, well... not maybe it
is not so minority.
On the other hand, you can't go wrong watching Shakespere or Baywatch!
Alexander the Other One
Alex Vinogradov wrote:
Hi Jim,Could you elaborate on that a little bit? I've justspent a bit (more
that I should have) of time workingwith a walk-forward optimization protocol,
only toconclude that it produces little more than confusion. What are your
thoughts?Thanks in advance,
Alex VinogradovMiami, FL
--- Jim Bronke wrote:>
It is better to include new relationships than to> walk forward old ones:
> > Jim Bronke> Phoenix, AZ>
> > > ----- Original Message ----- > From: "David Folster" > To: > Sent:
Sunday, January 13, 2002 9:06 PM> Subject: Walk forward optimization> > > :
Anyone have an opinion, good or bad, on walk> forward optimization?> : > :
Thanks,> : > : David> : > : > >
|