|
PureBytes Links
Trading Reference Links
|
----- Message d'origine -----
De : Gwenael Gautier <ggautier@xxxxxxxxxxx>
À : <omega-list@xxxxxxxxxx>
Envoyé : mercredi 22 septembre 1999 09:19
Objet : AW: Neurofuzzy, ex-Trailing stops are invalid
>
> | Maybe yes, maybe no.
> | If I succeed to build trading systems that are better than 4* DD in 200
> | weeks,
> | I'm probably not the only one.
> |
>
> 4*DD in any 200 trades, not weeks, unless it takes 200 or more weeks to do 200
> trades.
Easier in this case...
I have never seen, for a system that I have backtested and kept as a valid one,
the case where the max DD was above 2 times the max DD during historical training
or building period.
I usually backtest systems on thousands of bars ( 50,000 at least) and this produces
not only 200 trades but rather 2000 or 3000.
And again, the system that passed the test were considered as valid.
None has of course, after, reached the 2*DD limit ( actualized to the price level:
A max DD of 300 points occuring in a trading range around 1500 has not
the same meaning when it occurs in a 500 price area and the reverse)
>
> So let's start right here:
> 1- What is your trading philosophy then?
I have never traded without a system and will never do it
I let the computer build the system and test for me.
I do not code anymore, because I know of my limits after 10 years.
I'm stupid ,dumb and slow as possible when compared to the neurofuzzy thing.
> 2- Why do you think it procures an edge?
Statistics.
Only statistics.
A system that as learned over 2000 bars and that continues to work over 100,000 bars
means that something has been understood by the system.
> 3- How do you implement it?
See point 1
The only work that I keep for me is to choose the indicators and the training periods
> 4- Why do you think it neuro thing is not curve fitting?
The neurofuzzy thing IS curve fitting.
Anything in technical analysis is curve fitting, because the magic equation has never
been published ( or maybe out of print).
So, when you develop a method, a system, whatever it could be, you proceed by curve fitting
(using eyes, computer program, charts and so on), but it's always curve fitting because you have
to attempt to stick an ideal behaviour of the market ( buy -sell sequences , expected as perfect)
over sometyhing that is not perfect ( the market dataserie).
But there is a difference between intelligent curve fitting yelding to
some generalization and curve fitting for curve fitting ( overfitting if you want).
The test on unseen data ( I insist on a huge data set, mandatory ) is where you get the answer.
If it was really curve fitting under the common sense, no generalization can take place
on unseen data.
This also happens with neurofuzzy systems, and those falling into this case
must know where they will end thgeir short career: The system dustbin.
> 5- What makes you trust your signals signal after signal.
Statistics statistics and statistics.
You do not have to trust your signals after .
You have to trust yous system before.After, it's too late.
I also never trade only one system, but a minimum of 4.
So, the general and intermediate DD are reduced, and it avoids
to ask some questions when trading them realtime.
>
> Thank you in advance,
> Openly,
> Gwenn
Sincerely,
-Pierre Orphelin
web: http://www.sirtrade.com
|