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

AW: More Fun With Filters / don't pass on bandpass



PureBytes Links

Trading Reference Links

Hi Bilo,

Thanks for this interesting post, I am also heavily into cycles for
daytrading ES.

Do you really think that JMA is a linear filter? Thought it's nonlinear &
therefore not so good for cycle based approaches...

I think what most has been written about cycle analysis (& also what is sold
& promoted) is deeply flawed, because do have very seldom a clean DC & most
of the time there are many different cycles present causing translations
which makes cycle based trading difficult. However there is some practical
"light" that works in my eyes, most cycles are linked via some more or less
harmonic factors & you can take advantage of that when trying to build a
cyclic model....

I find cycle trading much easier when referring to multiple timeframes &
taking small scalps / daytrades in direction of longer term cycles that can
also be estimated without noise problems. 

One word about Ehlers & his approach: to split the market in trend & cycle
mode is nice from the theoretical side, but does not add much practical use,
what he calls "trend" is just the next longer term cycle...what is good is
his SNR that makes sure the DC has enough amplitude to be traded against the
noise...

If you use true FIR filters that notch out the DC like good old MA (very
good results with min effort, if used correct) you can find very "clever"
profit targets, the idea is based on Hurst's stuff & can be also found in
Kaufmanns great book, Trading Systems & Methods...

Sorry to be short, but its kinda late here & I need some rest :-))

All my best,

cosmic
 
  

-----Ursprüngliche Nachricht-----
Von: Bilo Selhi [mailto:biloselhi@xxxxxxxxxxx] 
Gesendet: Friday, July 09, 2004 6:24 AM
An: Omega List
Betreff: Re: More Fun With Filters / don't pass on bandpass

been following the thread,
going back to the "better than macd bandpass" topic,
the best simple bandpass i could find was, no surprise,
jma velocity which is a FIR bandpass filter. it has one 
important property that makes it valuable which is 
the length parameter it will take = dominant cycle length 
which simplifies buy and sell signal generation by a lot.

because of that jma velocity bandpass i found was 
near best as a tunable  FIR bandpass
filter, i even measured its freq response if 
anybody's interested, which is not perfect by far.
lag in this case is not the problem at all but the main 
problem is dcl measurement.

so i was wondering if anybody played with 
the cyclical system concept based on 
adaptive tunable bandpass ( tuning bandpass into dcl )or 
multiple bandpass bank( switching banks based on dcl )? 

i did not have enough oomph to finish 
the former again primarily because could not find 
a reliable way to forecast dominant cycle 
length ( more of a measurement problem ) 
but i have tried the latter but not with a great 
success either, again because of the dominant 
cycle measurement problem.
note: dcl is the main input into any bandpass model.

the consensus is that anytime your work with bandpass
based systems you get stopped out on dominant cycle 
measurement/forecasting, ie freq-time representation.
cycle is either don't persist enough to extract profits 
after measurement lag ( measurement lag problem ) or 
a cycle is simply not stable. while there are many known
ways to measure dcl, mesa, fft, cbf(jurik), hilbert, etc...
they are all  measurement methods which i found not reliable
( either too much/little lag and to small/large error ) 
literature on ft forecasting is virtually non-existent.

the question is then what's a good way to measure 
dominant cycle length with optimal lag and error?
how to minimize measurement lag and possibility 
or forecasting near term dominant cycle length?
comments?

all i wanted to say.
bilo.