|
PureBytes Links
Trading Reference Links
|
another follow up on the filter testing.
per MB, the aama is disigned to take care of the choppy
price series. not sure what Mark means by chop but
i assume chop is not just noise but simply sideways
chop? infering from here, sideways chop is simply:
signal = false, noise = true. and the task is to filter out
chop by making line smooth when price moves sideways ( no signal ).
if chop is not only sideways chop but noise overlayed on
vertical as well as horizontal price moves, than chop ain't chop,
it simply is noise.
anyhow, chop ( noise ) will have the following quantities that
will affect the output curve:
- chop shape
- chop size ( magnitude )
- chop duration ( frq )
- chop to signal ratio based on shape, size, duration...
plus on top of that signal quantities will also affect the chop filtering
process.
typically when studying filter you want to know these:
- frq. response curve
- step response
- impluse response
since aama is double adaptive filter then it should be able to adapt to
changing chop characteristics relative to changing signal characteristics
without the change in user inputs.
a good visual test would then be ( aside from basic step, impluse and common
periodic wave tests ):
- construct a mixed shape-ampl-duraiton chop wave
- overlay it on mixed shape-ampl-duration signal wave
lag studies would be easier to conduct under controlled conditions.
if you know when the chop starts/ends you can see the lag pretty well, etc...
mathwise i would infer that to be able to design a filter like that you need to
measure chop characteristics and feed them as filter controls, also
some sort of memory decay must be used as chop for last 30 bars ain't
same a chop for last 1000 bars... ie adaptation. so the filter should be able
to look back and differenciate between chop and signal, take recent chop
and adapt filter to that only, typical adaptive filtering.
bilo.
----- Original Message -----
From: Bilo Selhi
To: omega-list@xxxxxxxxxx
Sent: Monday, June 03, 2002 3:37 PM
Subject: Re: Re[2]: MB's AAMA vs OddBall, any comments?
as long as the black box ama as a
filter has characteristics that can be measured
in some way and as long as the descriptive math is
known without the details , it can be used even as a black box.
( gray box). actually knowing math is optional.
an example would be this:
- i tell an engineer to design a filter with the following specifications a,b,c...
- he designs it for me and gives the all basic filter characteristics that match
specifications.
in this case i don't really care about the details and the exact code
behind the filter.
all i know it works and it has certain quantities ( characteristics ) as
per my specifications.
if however i don't know anything about the filter, then i have an option
of studying the black box, running it through a filter toolbox for example
and finding out it's basic characteristic. this way you can hack the filter by
studying outputs per given inputs, comparison, etc...
that's what Mark asked everybody to do, i think. most likely Mark
wanted everybody to pitch in on studying the filter characteristics
by comparison.
however many, don't know how to do it properly.
you need to study the filter under controlled conditions.
this is done by running a known time series through a filter
and looking at the output.
known time series - filter - output curve
so you have the generate generic time series like, sine, sawtooth,
rectangle, rampup waves and run those through the filter to
know what its response characteristics are.
testing on unknown time series like price-time series for any symbol
is not a controlled environment and it's much harder to know how
the filter behaves.
i have a filter toolbox that generates various waves that could be
used to study any filter, i used it to study JMA.
if Colin west allows i can upload it to www.traders2traders.com
chop( noise) response can be studied easily by creating
choppy vs smooth price-time series of any form.
bilo.
----- Original Message -----
From: Mark Brown
To: omega-list@xxxxxxxxxx
Sent: Monday, June 03, 2002 1:17 PM
Subject: Re[2]: MB's AAMA vs OddBall, any comments?
Hello Robert,
RL> For two reasons I did not do much testing with it.
RL> One reason being lack of time but the other reason is that I don't
RL> know where this will lead.
RL> What are Mark's intentions with this AAMA?
i asked those who had the time and wanted to of own free will to
simply see how the aama stacked up against other popular versions.
from that response i made the last release #3 which should be the kill
all ama. if not i need to go back to work. i am only concentrating on
securing the truth about how good it is right now. i have not thought
beyond that - because it is no use if it does not perform.
RL> Is it for demo only? for co-testing? for sale?
it is a demo only at this stage, i would sell the source for 100,000
so yea i guess it's for sale if you want it. call br 549 to place
order.
RL> What to do with yet an other black box?
nothing - be quit and go away if you care not to help.
RL> My guess is that MB just wanted some people to co-test the value
RL> of this AAMA (which if fine by itself).
good then shut your flap. i now regret making a system out of it. i
should have just left it a pretty indicator so it could obtain cult
status someday when i run ads in magazines and talk of star warzie
math code. and to those who would not trust anything blackbox - if i
cracked it open you would not know what you were looking at anyway.
that's about as stupid as saying i will not fly in any air craft
unless i personally build the engine. bullcrap. it's greed - pure and
simple go make your own. but if your money motivated you don't have to
make your own shovels to become a millionaire digging ditches. if you
want to though thats your prerogative.
RL> The code is well protected and also limited by date/password.
RL> Hence anything I would do with it would be a waste of time since
RL> it would stop working the next month.
you have my opinion by now.
RL> While I would gladly run it with other type's of MA's for a
RL> testspin, I know other people already did this.
RL> Just some thoughts,
RL> Robert
--
Best regards,
Mark mailto:markbrown@xxxxxxxxxxxxx
|