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

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



PureBytes Links

Trading Reference Links




<SPAN 
class=033543211-02102003>yuki, what you're saying would seem to hold true for 
pretty much any trend-following indicator, what alternatives do you see as 
profitable in the long run?
<SPAN 
class=033543211-02102003> 
<SPAN 
class=033543211-02102003>dave
<BLOCKQUOTE 
>Hi 
  J,Thursday, October 2, 2003, 7:18:49 AM, you wrote:e> yes 
  it makes sense. I used a ema that worked well but their neare> values 
  did not....and when I backtested it with new data thee> results were 
  not good.. I always use half of the historical datae> to adjust system 
  (parameters, risk managemnt, etc)..Once I finishe> (before money 
  management) I ran a backtest and see how would havee> done in the 
  "future" (with the other 50% percent of historicale> data). Also I 
  backtest contracts seprately ...in this way you aree> sampling data and 
  seeing if returns are consistent...another waye> would be monthy 
  testings.e> BTW, I am still looking for a robust system with emas 
  cross....How?e> 1) filter noise - waveletse> 2) adaptive 
  parameters (voli adjusted, pattern adjusted, etc)e> 3) neural 
  networks?e> Any comments?I have serious doubts that it can 
  be done successfully over the longterm (MA crossover systems). It's 
  generally accepted that marketstrend less often than they chop. With any 
  kind of MA crossover,you're going to suffer some amount of lag.  When 
  you do catch atrending period, you may outperform for some period of 
  time.  Buteventually the chop will catch up to you I suspect.  
  Moreover, yourtrend needs to be so clean that you don't get taken out of 
  it on whatis simply a correction. Set your MAs (or EMAs, or whatever) 
  tootightly, and welcome to 'Whipsaw World'.  Set them too 
  loosely(filtering "wavelets, which in my mind can only be defined after 
  thefact), and you'll get taken out at the bottom of corrections and 
  lotsof other cycles, and put back in just as the cycle is ready to 
  rollover on you again. I have never seen a MA crossover that 
  worksconsistently over the long run. In the short run, and granted 
  theshort run can persist for some time, they can seem to be great.  
  Butthen comes payback time.It all boils down to one question: Can 
  you make enough during "clean"trending periods to pay for the high cost of 
  chop? Somebody may havefigured out a way to do that, and I can do it quite 
  well over shortand selected periods of time.  But I cannot do it over 
  the long termwith MA crossovers, and my experience is that the chop will 
  come, andit will wipe out the gains from the trends.The age-old 
  dilemma:  The longer the MAs, the less noise, but theharder the lag 
  is on your account.  The shorter the MAs, lag isn't aproblem, but the 
  noise is hard on your account.Yuki






Yahoo! Groups Sponsor


  ADVERTISEMENT<img
src="" width=300 height=250 border=0>  









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 the Yahoo! Terms of Service.