|
PureBytes Links
Trading Reference Links
|
Dimitris,
You are correct in that the last 71 bars are not "real". They can't
possibly be.
Whether they are useless or not is another question. I fill them in
with a projection from a parabolic fit of three data points or by
taking the data from the previous "cycle". The problem then falls to
deciding how to best use the cyclic information that the bands
present. For instance you can look at the prior touches of the bands
and count the time between touches and make a judgement as to
whether or not the current position within the DE bands makes sense
in context with the expected cycle. One could further add an
oscillator to help confirm the conclusion from the simple method I
described above.
If you have read Hurst's work you'll see that they are a convenient
way to observe and quantify the cyclic action. However, even Hurst
cannot have possibly found a way to project them into the future
with absolute certainty. It simply can't be done without the use of
voodoo or magic. You just can't know tomorrow for sure before it
happens.
However, I don't believe the information is completely useless when
taken in context along with other observations and/or indicators. In
other words it becomes a subjective technique much like 99% of all
TA work. Some systems can be quantified in terms of performance
statistics, some cannot as they are due to either skill of
interpretation or simply don't lend themselves to a consistent
mechanical system.
Anyone can modify this code to what they would like. Null is fine
past the end of the CMA if that's what you prefer. I simply showed
the projection routines that I found easiest to program with AFL and
that seemed to show possible future paths that were not
unreasonable. I make no claim regarding the accuracy of the
projection. In fact I'm pretty sure its not all that accurate.
Fred suggested recording basically updating the LastValue and
recording it daily thus preserving today's value of the endpoint for
each day. I've done this using a different program - something
custom that I created before I had Amibroker - and the resulting
indicator really doesn't look pretty. THe bands can be all over the
map. That makes does suggest that the parabolic fit isn't 'correct',
but we already knew that. I can only say I wish it were that simple!
In my opinion cycles are a subjective technique with continuous
variability. So its subjective and interpretive at best. There
seemed to be a call for something like SIgma bands and this was my
attempt. I'm more of a breakout trader anyhow and I don't use the
bands too much. I'll look at them as a point of interest
occasionally. If anyone can improve upon and test the reliability
please let me know. I stopped chasing it awhile ago.
-ace
--- In amibroker@xxxxxxxxxxxxxxx, "DIMITRIS TSOKAKIS"
<TSOKAKIS@xxxx> wrote:
> Ace,
> You use, for the CMA3 for example, the
>
> p3=144;
> Ave3=MA(Avg,p3);
> Lag3=(p3+1)/2;
> // Center the MA's
> cma3=Ref(Ave3,Lag3);
>
> The formula is wrong for the last 71 bars.
> Any further use of these last 71 bars will cause troubles to the
next
> steps. Since you make a multiple use of Ref(X,a), a>0, it is not
easy
> to find the extension of the wrong results.
> Unfortunately the AFL point of view for the last 71 bars is that
they
> are equal to the 72th [from the end] existing bar.
> Is it true ? Nobody knows, since they belong to the future.
> Should these 71 bars be considered as equal to the 72th bar ?
> IMO there is no reason or evidence.
> It is obvious now that ANY further use of these 71 bars [try to
RSI
> them, for example] will give some results [you may RSI anything,
from
> the math point of view...] but these results will be improper for
use.
> Dimitris Tsokakis
>
> --- In amibroker@xxxxxxxxxxxxxxx, "acesheet" <acesheet@xxxx> wrote:
> > Fred,
> >
> > >>> Doesn't the centerline change every time a new bar of data
is
> > added ?
> >
> > Yes. For each bar that's added due to progression in time the
> > centerline gets updated from being extrapolated to being actual
> > data. When I said the CMA never changes what I meant was its
always
> > calculated from the 'real' data before the triangles and I never
> > modify it. Its always a shifted SMA. It only becomes bogus
> > (projected data) when you get into the time frame to the right
of
> > the triangles.
> >
> > >>> For example in a 250 bar CMA when a new bar is added doesn't
> the
> > point for the center line that is now 126 bars ago change from
> being
> > extrapolated to being calculated ? and doesn't that in turn
affect
> > the extrapolation afterwards and also the bands as well ?
> >
> > It absolutely effects the extrapolation and the band width. That
> was
> > my point. That's why you see the changes in the indicator from
day
> > to day. Its definitely a problem, but unless you can predict the
> > future with certainty you need some method of extrapolation and
all
> > methods of extrapolation will suffer from the same problems to
one
> > extent or another. It cannot be avoided. That's why I said
Maggio
> > must being doing something similar. He may have a better method
of
> > extrapolation or curve fitting, or maybe not. It could just look
> > pretty. I really don't know. Since his is a 'black box' and
nobody
> > body he knows how it works, it makes it rather difficult for me
to
> > trust the pictures that are presented. But I'm an engineer and
so
> > also a scientist and I nned to understand how it works before I
use
> > it. Its a blessing and a curse! :)
> >
> > Hopefully I answered you questions adequately.
> >
> > -ace
> >
> > --- In amibroker@xxxxxxxxxxxxxxx, "Fred" <ftonetti@xxxx> wrote:
> > > Doesn't the centerline change every time a new bar of data is
> > added ?
> > >
> > > For example in a 250 bar CMA when a new bar is added doesn't
the
> > > point for the center line that is now 126 bars ago change from
> > being
> > > extrapolated to being calculated ? and doesn't that in turn
> > affect
> > > the extrapolation afterwards and also the bands as well ?
> > >
> > > Again I haven't looked at your code in detail but this is of
> > course
> > > the issue that Dimitris was pointing at as well.
> > >
> > > --- In amibroker@xxxxxxxxxxxxxxx, "acesheet" <acesheet@xxxx>
> wrote:
> > > > Oh, a couple of other things I forgot to mention about the
DE
> > code
> > > I
> > > > presented.
> > > >
> > > > 1) The triangles ARE where the centered MA stops and the
data
> > > > projection begins.
> > > > 2) The centered MA never changes, just the band width and
the
> > > > projection of the centerline.
> > > >
> > > > -ace
> > > >
> > > > --- In amibroker@xxxxxxxxxxxxxxx, "acesheet" <acesheet@xxxx>
> > wrote:
> > > > > Here's a little explanation.
> > > > >
> > > > > The reason the band values change for future dat in the
code
> > that
> > > > I
> > > > > presented is that the widths are calculated based on 2
> > standard
> > > > > deviations from the mean.
> > > > >
> > > > > I believe this to be valid because the act of centering
the
> MA
> > > > makes
> > > > > the distribution about the centered moving average a true
> > average
> > > > of
> > > > > the data valid to the bar period/2. Therefore, since the
> > > > > mathematical operation is an average, and if we assume
some
> > cycle
> > > > is
> > > > > currently active in the form of a sinusoid, then the
> > distribution
> > > > > should be close to a normal distribution if enough data
> points
> > > are
> > > > > taken, so standard deviation is the proper way to describe
> the
> > > > > envelope width if its a normal distribution.
> > > > >
> > > > > In fact statisticians call standard deviation with the
greek
> > > > letter
> > > > > sigma - ala sigma bands - so that's probably why Maggio
calls
> > his
> > > > > bands 'sigma bands'.
> > > > >
> > > > > Here's the calculation for the bandwidth in terms of
percent:
> > > > >
> > > > > dp1=LastValue(Ref(StDev( (Close-cma1)/cma1, k*p1 ),
(p1+1)/2-
> > 1));
> > > > >
> > > > > where k=3 means that the standard dev is calculated over
3x
> > the
> > > > > period number of data points.
> > > > >
> > > > > Using Lastvalue() makes the width a constant for all time
> > based
> > > on
> > > > > today's calculation. Therefore as data is added the
> bandwdiths
> > > > > change and the way I have it coded it changes for all
time.
> > Its
> > > > > almost a derivative of Bollinger bands. Bollinger was on
the
> > > right
> > > > > track he's just using the average incorrectly and standard
> > > > deviation
> > > > > does not describe the price distribution about a non-
centered
> > SMA
> > > > > which is why his bands vary in width with the volatility
of
> > the
> > > > > stock price movements.
> > > > >
> > > > > I will bet you almost anything Maggio's charts do roughly
the
> > > same
> > > > > thing with whatever smoother or centerline calculation
filter
> > > he's
> > > > > using. There are many ways to calculate the centerline.
His
> > chart
> > > > > could simply be a smoothed and centered SMA. You could
easily
> > > > smooth
> > > > > a CMA with a parabolic curve fit or with various forms of
> > > > > regression. However, he still needs to project the bands
> > forward
> > > > in
> > > > > time using some method.
> > > > >
> > > > > If you are really amibitious you should investigate least
> > squares
> > > > > regression fitting of trigonometric functions. I can
create
> > > > > beautiful DE bands with them using a different program
that I
> > > > wrote.
> > > > > You still need to project forward in time, however, so
that's
> > > > really
> > > > > the rub. There are likely much better ways of projecting
the
> > line
> > > > > forward than I presented here using various regression
> > > techniques.
> > > > > Learning something about digital signal processing is
> probably
> > > > also
> > > > > key.
> > > > >
> > > > > In evaluating Maggio's service ask yourself a few
questions.
> > Why
> > > > > create a for pay website to sell the idea to others if it
> > works
> > > so
> > > > > well? Why not just trade your way to financial freedom and
> > > retire?
> > > > > Beware of snake oil. Maybe he's got something good there -
I
> > have
> > > > to
> > > > > admit it looks good - but I sure don't blindly trust a web
> > site.
> > > > >
> > > > > I'm sure a discretionary trading system could be built
using
> > > > > envelopes and oscillators that would perform fairly well,
> > however
> > > > > there would really be no way to backtest it. Forward
testing
> > or
> > > > > trading would be the only way to do it.
> > > > >
> > > > > -ace
> > > > >
> > > > > --- In amibroker@xxxxxxxxxxxxxxx, "Fred" <ftonetti@xxxx>
> wrote:
> > > > > > Dimitris / WaveMechanic,
> > > > > >
> > > > > > If the length of the CMA is n then
> > > > > >
> > > > > > The CMA can only be calculated up to n/2 bars ago after
> > which
> > > it
> > > > > must
> > > > > > be extrapolated via some technique.
> > > > > >
> > > > > > So for example if one wanted to plot a 250 bar CMA
showing
> > the
> > > > > > history of where it had been at time of original
> calculation
> > > > then
> > > > > one
> > > > > > would need to,
> > > > > >
> > > > > > - At bar 375 calculate the CMA for bars 1 through 250
and
> > > > > extrapolate
> > > > > > for bars 251 through 375. This would provide the
initial
> > 250
> > > > > > plottable points.
> > > > > >
> > > > > > - At bar 376 calculate the CMA for bars 2 through 251
and
> > > > > extrapolate
> > > > > > for bars 252 through 376. this should add one and only
one
> > > > > additional
> > > > > > plottable point i.e. the one at bar 376.
> > > > > >
> > > > > > - This process could then continue up through the
current
> > bar.
> > > > > >
> > > > > > Someone made mention of Jan Arps Sigma Bands code for
> > > > TradeStation
> > > > > > which although available for usage in TS is not
viewable.
> > It
> > > > does
> > > > > > however provide the capabilty of showing both the
> historical
> > > > past
> > > > > > datapoints as originally calculated as well as the
current
> > > > > picture.
> > > > > > The only similarity between these would be the CMA for
the
> > > > cureent
> > > > > > and any future bar.
> > > > > >
> > > > > > See his description here ...
> > > > http://www.janarps.com/SigmaBands.htm
> > > > > >
> > > > > >
> > > > > >
> > > > > > --- In amibroker@xxxxxxxxxxxxxxx, "wavemechanic"
> <wd78@xxxx>
> > > > wrote:
> > > > > > >
> > > > > > > ----- Original Message -----
> > > > > > > From: DIMITRIS TSOKAKIS
> > > > > > > To: amibroker@xxxxxxxxxxxxxxx
> > > > > > > Sent: Saturday, February 21, 2004 3:18 AM
> > > > > > > Subject: [amibroker] Re: Sigma Bands
> > > > > > >
> > > > > > >
> > > > > > > Wayne,
> > > > > > > Sorry, I can not agree that the history of the
signals
> > is
> > > > > > meaningless.
> > > > > > > I need always to check any trading idea from its
past
> > > > behavior.
> > > > > > >
> > > > > > > Spoken like a true system trader. However, looking
at
> > > > > Maggio's
> > > > > > description of the bands it is not clear to me that
there
> > > should
> > > > > be
> > > > > > any change as new data is added, reflecting the fact
that
> > the
> > > > > bands
> > > > > > are simply the sigma of the % change of the data from
the
> > CMA.
> > > > > This
> > > > > > is a bar by bar calculation that does not change as new
> data
> > is
> > > > > > added. Perhaps, as you suggest, the problem lies in the
> way
> > > CMA
> > > > > is
> > > > > > calculated by the code. If so there must be a way
around
> > this
> > > > > > problem, as evidenced by the fact that a manual
calculation
> > of
> > > > CMA
> > > > > > does not look into the future but simply centers a MA
> within
> > > the
> > > > > > incremental period of another MA. So one needs to make
> code
> > > > > > duplicate the manual calculation which is
straightforward.
> > > Does
> > > > > not
> > > > > > sound like rocket science. And extrapolation of the CMA
> > does
> > > > not
> > > > > > change its previously established values. However, even
> > when
> > > > > things
> > > > > > are working right neither Sigma Bands or Hurst Channels
by
> > > > > themselves
> > > > > > provide a mechanical buy/sell signal. No problem for
> > > > > discretionary
> > > > > > traders but system traders will need some "antacid" in
> order
> > to
> > > > > avoid
> > > > > > heartburn. LOL.
> > > > > > >
> > > > > > > Dimitris Tsokakis
> > > > > > > > Dimitris,
> > > > > > > >
> > > > > > > > I think you don't get that 'signals of the past'
are
> > in
> > > > the
> > > > > > past.
> > > > > > > It's
> > > > > > > > history and as such is meaningless. Hurst
developed
> > his
> > > > work
> > > > > > before
> > > > > > > > computers had the power to do billions of
> computations
> > > per
> > > > > > second.
> > > > > > > > Therefore, the idea of backtesting his work is a
> waste
> > of
> > > > > time.
> > > > > > > >
> > > > > > > > The Sigma Bands we are discussing seem to be a
> > derivative
> > > > of
> > > > > > the
> > > > > > > Hurst
> > > > > > > > Dependency Envelopes, so ably programmed by Ace...
> and
> > > > much
> > > > > > > appreciated
> > > > > > > > too. I feel there is nothing at all to be gained
from
> > 1)
> > > > > > looking
> > > > > > > into
> > > > > > > > the past or 2) trying to guess the future. I want
to
> > > know
> > > > > what
> > > > > > is
> > > > > > > > happening right now. The Sigma Bands MAY offer
some
> > > > insight
> > > > > > into
> > > > > > > what
> > > > > > > > the market is saying now, but should never be used
to
> > > > trade
> > > > > any
> > > > > > > market
> > > > > > > > by themselves. This information should always be
used
> > as
> > > > > > > confirmation of
> > > > > > > > other indicators and trading techniques.
> > > > > > > >
> > > > > > > > Computers will never replace the human brain.
> > > > > > > >
> > > > > > > > Wayne
> > > > > > >
> > > > > > >
> > > > > > >
> > > > > > > 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
> > > > > > > Yahoo! Groups Links
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
Yahoo! Groups Links
<*> To visit your group on the web, go to:
http://groups.yahoo.com/group/amibroker/
<*> To unsubscribe from this group, send an email to:
amibroker-unsubscribe@xxxxxxxxxxxxxxx
<*> Your use of Yahoo! Groups is subject to:
http://docs.yahoo.com/info/terms/
|