|
PureBytes Links
Trading Reference Links
|
Good morning Lords and Ladies of the List!
I swam in that lake about two years ago and no matter
how deep I dove, not a single damn sword anywhere to
be found. Now this could have been the lack of purity
of my test systems but...
I spend a few thousand dollars for a "basket of
systems" from a vendor whose name I will not mention
in a broadcast email (though private inquires are
always welcome). This Dark Knight advertised his
systems saying they worked great using walk-forward
optimization. Of course, TS does not do this easily so
there was no simple test method to validate his claim
to a seat at the Round Table.
I, Alexander the Tenacious, performed hundreds of
tests with two computers over a number of these
systems with adjusting both the optimization periods
and the walk-forward trading periods. On a daily
chart, I tested periods from two weeks, very high
frequency optimization and trading cycles to five
years or very low frequency optimization and trading
cycles.
I found that across all systems (I tested five), the
best results came from an average of two year
optimization with an average four year trading cycle
before the next optimization. In other words, a little
optimization traded over a longer time was better.
Overall however, the trading results, to use an
ancient noble English term, sucked!
Curve fitting works great on Halle Berry and Cameron
Diaz. Not on time series data.
I hope this helps!
alexander the anecdotal
--- CRLeBeau@xxxxxxx wrote:
> Michael wrote> This brings to mind what I consider
> a vital question
> regarding
> re-optimization. Does anyone have any hard cold
> statistics as to the
> workability of "walk forward" (or re-optimization)
> of systems? Or do you
> know where I can find the information? I've seen and
> remember debates but
> don't remember any black and white comparisons or
> any actual trades in any
> of the discussions.
> * * * *
> Michael re hard evidence
>
> Louis B. Lukac and B.Wade Brorsen wrote a graduate
> thesis entitled "The
> Usefulness of Historical Data in Selecting
> Parameters for Technical Trading
> Systems". Their study compared the results of fixed
> parameters and periodic
> reoptimizations of the parameters on two
> trend-following systems.
>
> They concluded "The results of all the tests suggest
> that the forcasting
> ability of optimization is limited. Optimization
> was not able to forcast
> parameter sets which would produce portfolio profits
> better than a random
> selection strategy."
>
> This informative study was reprinted many years ago
> by Ed Dobson at Traders
> Press. (he sold it for about $10. I don't know if
> it is still in print).
> As I recall the best results were obtained by
> finding one good set of
> parameters and then sticking with those. Periodic
> reoptimizations did not
> improve the test results.
>
> This is only one study but I have never seen any
> study that showed that
> reoptimization actually helped. However if you give
> any value to anecdotal
> evidence there are many traders that seem to be fond
> of this questionable
> technique.
>
> Chuck LeBeau
> www.traderclub.com
>
|