|
PureBytes Links
Trading Reference Links
|
rudolf stricker wrote:
> Playing around with some SystemTests I wonder, which optimization
> criterions might be the "best":
>
> Standard "profit - loss" delivers very big "local" increments in
> the equity distribution (for a five years period), at least for my
> data.
>
> What about "profit / loss" ? Does this make sense?
> What about an (eg exponential) increasing "rating" vs time to
> concentrate on the "newest" data?
> What other optimization criteria may be useful?
> Any hints?
>
> mfg rudolf stricker
> | Disclaimer: The views of this user are strictly his own.
Hi Rudolf,
After giving its (pretty long and disappointing) due to the idea of optimization of a trading
system, I got convinced that an ultimate criterion for a good system is a "quality" of its
equity line. Ideally, it should be trending up with minor retracements. If you consider as
an example of a good (in fact, too good) system the one with 1/1 as #win/#loss ratio, and
2/1 as the average ($) win/average($) loss ratio, you may see this kind of the equityline
being produced. Certainly, unless you have nerves of steel and a sleep of baby, you may
want to minimize drawdowns as well, or # of losses in a row. It would make the line even
less susceptible to steep/long downs. To put it into a quantitative framework, one could
use RSI applied to the equity line, and ask for a relatively high value as averaged over a
test period, or any other, similar variations on the theme. Unfortunately, it may be hard to
do in Metastock, since you will have to program the equity line calculation.
Mind what I am talking about is not searching for a system with the largest gain, but with
highest "consistency" in gaining, otherwise it would be more about gambling, IMHO.
All this is just a feeeling I've got playing with optimization ideas, so no hard proofs ...
Cheers, Vitaly
|