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

[amibroker] Re: Auto-optimization AFL uploaded



PureBytes Links

Trading Reference Links

Dave,

IMHO more information is better than less information.

I understand the probabilities of success a lot better when instead 
of picking an arbitrary fixed length of time to perform rolling 
optimizations over especially when those periods do not cover at 
least bull and bear segments one picks a fairly long series of data 
that contain hopefully at least several bull and bear segments and 
optimizes a system over it and then tests out of sample to see that 
it performs well on at least the next bull and bear segement and then 
reoptimizes the system on all the data i.e. what was previosly in and 
out of sample and then begins to trade the system and then 
periodically reoptimizes the system with what is at that time all the 
data although the reoptimzation process should not be required  
frequently.

--- In amibroker@xxxxxxxxxxxxxxx, "Dave Merrill" <dmerrill@xxxx> 
wrote:
> I'm really tempted to conclude that myself, since very little I've 
found so
> far performs very well like this, but...
> 
> - what's so different about "re-optimizing" than optimizing once? 
why are
> the settings you get optimizing up to today more useful than what 
the guy
> gets who does the same thing a year later?
> 
> - if we don't optimize at all, where do "standard" parameter 
settings come
> from? books? newsgroups? seminars? urban legends? where do these 
ideas
> originate, if not from someone optimizing or testing something at 
some
> point?
> 
> - when you're backtesting different methods, whether they're 
trading rules
> or settings, success depends on being able to tell something about 
future
> performance from past performance. if you can't do that, 
reoptimizing won't
> work, but no kind of backtest will be useful either. how do we 
design and
> test trading systems at all if that's the case?
> 
> - if you can tell something about the future from the past, why not 
continue
> learning about market behavior after trading starts, which is what
> reoptimizing does? why would it improve a system to ignore possible 
changes
> in market behavior after that? wouldn't you at least want to see if 
you're
> still making money, and possibly adjust, or stop trading until you 
got a
> better idea? how is this fundamentally different from reoptimizing?
> 
> - maybe the problem is that changing settings too often hurts 
performance.
> the framework lets you optimize only as often as you like, and if 
that
> really is the problem, performance should improve with more time in 
between.
> in my experience, nothing that simple happens.
>   even if optimizing every year, say, worked better, isn't it 
pretty random
> where during the year that falls? if settings end up the same 
whenever it
> falls, then they aren't changing often enough for reoptimizing more
> frequently to have hurt you. if they come out differently 
optimizing every
> june than every january, how do you justify choosing when to do it, 
or know
> that you won't end up wrong-headed for 11.999 months out of every 
year and
> lose your shirt?
> 
> 
> I'm not trying to be argumentative, honest. I just don't see any 
fundamental
> reason why reoptimizing shouldn't work that doesn't invalidate the 
whole
> idea of backtesting. I've thought about all this a great deal, and 
I really
> do wish somebody would help me think my way out of this box I've 
thunk
> myself into.
> 
> so far, my only possible specific theories are either that 
everything that
> appears to work really doesn't in the long run if you test it 
right, or the
> metrics we use to evaluate past performance don't predict the 
future very
> well.
> 
> there's also the possibility that no matter what this "logic" says, 
it's
> just plain wrong, for some reason I don't get. yet.
> 
> dave
> 
>   I have read nearly one meter of books and I remember one " the
>   ultimate trading guide". it backtests p 216 a simple donchian 
system
>   that is reoptimized on a rolling 2 years basis, the results were
>   worst than static parameters and the conclusion was that
>   reoptimization does not work...
> 
>   stephane


------------------------ Yahoo! Groups Sponsor ---------------------~-->
Rent DVDs from home.
Over 14,500 titles. Free Shipping
& No Late Fees. Try Netflix for FREE!
http://us.click.yahoo.com/ybSovB/hP.FAA/3jkFAA/GHeqlB/TM
---------------------------------------------------------------------~->

Send BUG REPORTS to bugs@xxxxxxxxxxxxx
Send SUGGESTIONS to suggest@xxxxxxxxxxxxx
-----------------------------------------
Post AmiQuote-related messages ONLY to: amiquote@xxxxxxxxxxxxxxx 
(Web page: http://groups.yahoo.com/group/amiquote/messages/)
--------------------------------------------
Check group FAQ at: http://groups.yahoo.com/group/amibroker/files/groupfaq.html 

Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/