[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Optimal f and system optimization


  • To: metastock@xxxxxxxxxxxxx
  • Subject: Re: Optimal f and system optimization
  • From: "rudolf stricker" <rst@xxxxxxxxxxx>
  • Date: Mon, 26 Jul 1999 12:46:54 -0700
  • In-reply-to: <37919bc4.8775155@xxxxxxxxxxx>

PureBytes Links

Trading Reference Links


Glen,

On Mon, 19 Jul 1999 10:34:40 -0700, you wrote:

>That sounds really interesting.  I'd like to hear what you find.

Using some kind of genetic algorithm for system optimization, I do not
get a single "best" solution but a bunch of systems with different
parameter sets but (nearly) the same result in terms of the
optimization function. Testing these systems with the Optimal f method
via post-processing (like described in Futures Magazine) gives rather
different results for many of these systems. 
My conclusion: f is *not* an independent parameter that can be
optimized separately. So f should be included into the list of
parameters for system optimization to take care of the inherent
dependencies.

>Be careful not to over-optimize, though.  Your primary goal should not be to
>maximize profits themselves, but rather to maximize *consistency* of
>profitability from your system.  From here, your money management system and
>time will maximize overall profit.

Over-optimization is a bad thing by definition, but
"under-optimization" is bad as well. Imo, all relevant system
parameters (including those for "money management") should be
optimized synchronously, if a parameter is not really independent from
all others. 

Concerning your criterion "variance or standard deviation
of returns" for consistency of profitability: I personally would
prefer to concentrate on "consistency of losses" to avoid any
unnecessary  restrictions for wins. And the
"probability-to-loose-more-than-xx%" (p2lmtxx5) may be a good
criterion.

Unfortunately, a reliable calculation of p2lmtxx% requires a valid
analytical approximation for the probability of losses, especially
when it comes to xx values of eg 50 ... 80 ... 90 ... %. So I'm still
interested to learn, what experienced people are using here ...

mfg rudolf stricker
| Disclaimer: The views of this user are strictly his own.