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

[amibroker] Re: Robustness testing & Over-optimization checklist



PureBytes Links

Trading Reference Links

Yes, that's exactly it.

What I was trying to get across is two things.

1. No pencil points.  They don't work moving forward.

2. Your optimization parameter vs. performance rating is essentially 
a probability distribution curve.  You would ideally want a curve 
where most (68%) of the values are not too far (1 std. dev) from the 
mean.

HB

--- In amibroker@xxxxxxxxxxxxxxx, "Gary A. Serkhoshian" 
<serkhoshian777@xxxx> wrote:
> HB,
>  
> To break this discussion down into it's simplest sense, are you 
essentially saying that our sensitivity analysis should show a 
smoothed curve extending from the optimal paramater set versus 
a "pencil point" (i.e. sharp drop off in performance as we move away 
from optimal params)?
>  
> I think the 68% is throwing people off (myself included).
>  
> Thanks,
> Gary
> 
> Dave Merrill <dmerrill@xxxx> wrote:
> somewhat, yes, it makes sense. not completely sure how to quantify 
this though. the range of values constituting a plateau, within which 
68% of values should be "good", should be how wide? not numerically 
how wide, but what's the principle? 
>  
> for example, SAR responds incredibly nonlinearly to parameter 
settings. there are some ranges of values where change makes no 
difference at all, some that are very sensitive, and some that 
produce nonsensical results. there's a recommended paramter range, 
but it's odd. what does that say about its robustness?
>  
> dave
> The 68% thing is essentially the plateau thing.   You're right, in 
> the end, you will only select one parameter setting.  However, for 
> example, if your test show that a 10 period MA works amazingly, but 
a 
> 9 and 11 period MA don't work, then choosing a 10 MA is not the 
best 
> way to go.  Just because it took you out of a few losses in your 
> historical backtest, it probably won't work well moving forward.
> 
> But if your 10 period MA is part of a plateau of "good" values 
> (plateau includes 7, 8, 9, 11, 12, 13), then it would be ok to use 
> 10.  And 7, 8, 9, 11, 12, and 13 would encompass 68% of your "good" 
> values.  
> 
> Make sense ?
> 
> HB
> 
> --- In amibroker@xxxxxxxxxxxxxxx, "Dave Merrill" <dmerrill@xxxx> 
> wrote:
> > thanks for the feedback.
> > 
> > I get that there should be plateaus, not spikey peaks, in the 
yield 
> map as
> > you change paramter settings. what I don't get is the 68% thing.
> > 
> > does it mean you should restrict the range of each parameter so 
> that at
> > least 68% of the settings within that range are profitable? 
> typically,
> > people settle on one specific set of settings they hope will work 
> well for
> > some range of issues over som range of time. how is the system 
more 
> or less
> > robust by restricting the range of parameters over which you 
tested 
> to
> > arrive at the ones you actually use?
> > 
> > dave
> >   For #10, Kaufman is refering to the notion that the majority 
> (68%) of the
> > range of parameters should be profitable.  In a map of parameter 
> values &
> > associated returns, you want to see a plateau instead of just 
> spikes.  Check
> > out the 3D optimization spreadsheet in the Files section.
> > 
> >   Also, the order of parameters is crucial because you should 
test 
> the most
> > importanat variables first, i.e. the ones that cause the largest 
> changes in
> > your results.  Focus on those first and then on the less 
important 
> ones
> > 
> >   Yes, the book is excellent, not just for the robustness 
section.  
> I
> > haven't read his other trading system book yet though so I don't 
> know if
> > there are similarities.
> > 
> >   HB
> 
> 
> 
> Send BUG REPORTS to bugs@xxxx
> Send SUGGESTIONS to suggest@xxxx
> -----------------------------------------
> 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 the Yahoo! Terms of 
Service. 
> 
> Yahoo! Groups SponsorADVERTISEMENT
> 
> Send BUG REPORTS to bugs@xxxx
> Send SUGGESTIONS to suggest@xxxx
> -----------------------------------------
> 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 the Yahoo! Terms of 
Service. 
> 
> 
> ---------------------------------
> Do you Yahoo!?
> The New Yahoo! Shopping - with improved product search


------------------------ Yahoo! Groups Sponsor ---------------------~-->
Buy Ink Cartridges or Refill Kits for your HP, Epson, Canon or Lexmark
Printer at MyInks.com. Free s/h on orders $50 or more to the US & Canada.
http://www.c1tracking.com/l.asp?cid=5511
http://us.click.yahoo.com/mOAaAA/3exGAA/qnsNAA/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/