MG,
Agree that Fourier
is maybe alright if you’re looking at the big picture. But the
restrictions on stationarity mean that it’s useless for short cycles on
small data sets. I’ve used it to find a suitable periodicity for a Centered
MA with fixed-width envelopes. Which is a tool that’s only any use for ‘eyeballing’
the major cycles since you have to extrapolate it to the right hand edge.
I’ve
found Ehler’s Cyber Cycle measurement to be very good for practical
purposes as he starts off with the assumption that, yes there may be lots of
cycles going on at any one time but there is only one dominant/tradeable cycle…and
the indicator attempts to measure that.
Andy
From:
equismetastock@xxxxxxxxxxxxxxx [mailto:equismetastock@xxxxxxxxxxxxxxx] On Behalf Of mgf_za_1999
Sent: Thursday, September 08, 2005
9:09 PM
To: equismetastock@xxxxxxxxxxxxxxx
Subject: [EquisMetaStock Group]
Re: Synthetic cycles
I haven't read the original,
just read about it, and even this was
ages ago, so I am not sure how to apply it, apart
from this fact -
when you smooth you introduce
autocorrelation. This alone has helped
me quite a bit on a number of occasions.
I *think* that it means that, what to you may look
like nice tradeable
cycles, may be the result of the sum of lots of
random noise and not
repeatable in future.
I've toyed for ages with Fourier stuff, but the
accuracy sucks to say
the least. It is nice to extract cycles and
study cycle lengths and
comment on the time it will take before the next
top or bottom occurs,
but the trick is that straight line! I would
try to get that with
other techniques and then use Fourier on the
difference between *other
technique's straight line* and the market.
Fourier/cycles is a long term technique and I've
found shorter term
techniques (that is the same in the limit) that
gives better results
and prefer these now. But Fourier still is a
nice tool to have handy
if you study e.g. commodities or sunspots or
something that has *true*
cycles in them. I wish I had time to
continue working on these things
as well, as I think you can in fact do well if you
use cycles, but you
have to go waaaay beyond Fourier I guess.
Wavelets and fractional
integration and these kind of things. Some
I've looked at in detail,
some are just ideas I know about. The
'problem' again is the long
term nature and the 'predicting' nature of
cycles. To look at a cycle
and say the market will peak in two months' time
will make you either
a visionary or very poor (I am starting to sound
like super) - rather
focus on do you buy or sell today when a moving
average is crossed or
not. More practical and to the point, and
less neck on the line.
Regards
MG Ferreira
TsaTsa EOD Programmer and trading model builder
http://www.ferra4models.com
http://fun.ferra4models.com
--- In equismetastock@xxxxxxxxxxxxxxx,
"Andy" <AndyDavidson@xxxx> wrote:
> Nice one Jose! I was thinking about trying to
do something like that
myself,
> but I'm sure that what took you 10 minutes
would have taken me the
best part
> of a day.Thanks for saving me the time!!
>
>
>
> Now, if you imagine two additional factors:
>
> 1) An
underlying trend represented by a straight upward/downward
> sloping line, or even a much longer-period
cycle, that represents the
> "fundamental" influence on a
market.
>
> 2) That the
periodicity of all the cycles is not constant but
has some
> minor variation about their nominal value.
>
> And you can see how, when lots of
"simple" stuff is added together
it can
> easily look very complicated and random! This
is the general gist of
what
> Hurst
was on about.
>
>
>
> MG, what's interesting to me is that if you
plot Jose's indicator
and look
> at the individual cycles, then zoom out the
chart view sufficiently,
all the
> small cycles start to look rather like white
noise.
>
>
>
> Maybe I misunderstand the academics(!) but if
its the smoothing
process that
> creates cycles where none existed before then
why can we look at an
> un-smoothed price chart and *see* tradeable
cycles? Why do chart
patterns
> occur so frequently? Are these purely random
in nature, even on the
> week/month time scale? Surely it depends on
what
time-scale/magnification
> level you're looking at.i.e. what you
actually define as a "trade
cycle"? In
> other words, sure there must be an element of
noise at the very high
> frequency end of the spectrum (which even
itself looks like cyclical
> behaviour). But as you go up orders of
magnitude that effect must
decrease
> dramatically and more predicatable influences
take over. These would
mainly
> result, I would think, from the
crowd-mentality of speculators.but I
don't
> want to get into an Elliot argument here
(although I'm sure I can
see 3 and
> 5-wave patterns in Jose's indicator)!! Then,
eventually, the
"fundamentals"
> start to have a distinct influence. At some
point on this scale, as
noise
> gives way to crowd behaviour and then to
fundamentals, you get a
point where
> efforts at modelling/predicting what's going
on start to bear fruit?
>
>
>
>
>
>
>
> _____
>
> From: equismetastock@xxxxxxxxxxxxxxx
[mailto:equismetastock@xxxxxxxxxxxxxxx]
> On Behalf Of mgf_za_1999
> Sent: Thursday, September 08, 2005 6:53 AM
> To: equismetastock@xxxxxxxxxxxxxxx
> Subject: [EquisMetaStock Group] Re: Synthetic
cycles
>
>
>
> About 100 years ago some guys did a study on
autocorrelation - just
> read about it, but it was senstational at the
time. It was Slutzky's
> "The summation of random causes as the
sources of cyclical processes",
> originally published in Russian.
Another guy, Yule, came to similar
> conclusions, so this thing is called the
Slutzky-Yule effect and is
> the reason why stock price tables in news
papers show the closing and
> not the average price for the day. It
is also why many people use
> data that has not been seasonally adjusted
rather than otherwise.
>
> The effect is a bit difficult to explain
without using technical
> jargon, but let me try. If you smooth a
series you create implicit
> autocorrelation - implicit cycles if you
want. When you smooth you
> can create apparent systematic effects simply
because you are
> smoothing rather than because there is some
underlying factor
> responsible for it. This is the
senstational bit! You can use random
> noise, smooth it, and generate nice looking,
systematic effects. What
> Slutzky did and what shocked the academic
world at the time was to
> mimic an actual trade cycle using only random
noise.
>
> Regards
> MG Ferreira
> TsaTsa EOD Programmer and trading model
builder
> http://www.ferra4models.com
> http://fun.ferra4models.com
>
> --- In equismetastock@xxxxxxxxxxxxxxx,
"Jose Silva" <josesilva22@xxxx>
> wrote:
> > A nice representation of what a whole
bunch of superimposed cycles
> > actually looks like, can also be seen
with this MetaStock indicator:
> >
> >
> > ================
> > Synthetic cycles
> > ================
> > ---8<--------------------
> >
> > { 5-wave synthetic cycles v1.0 }
> >
> > { CCopyright 2005 Jose Silva
> > For personal use only.
> > http://www.metastocktools.com }
> >
> > { User inputs }
> > plot:=Input("Cycles:
[1]Composite, [2]Individual",1,2,1);
> > Sval1:=Input("Sine 1
value",-720,720,2);
> > Sval2:=Input("Sine 2
value",-720,720,6);
> > Sval3:=Input("Sine 3
value",-720,720,20);
> > Sval4:=Input("Sine 4
value",-720,720,50);
> > Sval5:=Input("Sine 5
value",-720,720,160);
> >
> > { Sine components }
> > sine1:=Sin(Cum(Sval1));
> > sine2:=Sin(Cum(Sval2))*.5;
> > sine3:=Sin(Cum(Sval3))*.25;
> > sine4:=Sin(Cum(Sval4))*.125;
> > sine5:=Sin(Cum(Sval5))*.0625;
> >
> > { Composite Sine }
> >
composite:=sine1+sine2+sine3+sine4+sine5;
> >
> > { Plot in own window }
> > If(plot=1,composite,sine1);
> > If(plot=1,composite,sine2);
> > If(plot=1,composite,sine3);
> > If(plot=1,composite,sine4);
> > If(plot=1,composite,sine5)
> >
> > ---8<--------------------
> >
> >
> > jose '-)
> > http://www.metastocktools.com
> >
> >
> >
> >
> > --- In equismetastock@xxxxxxxxxxxxxxx,
"teclogeo" <teclogeo@xxxx>
> > wrote:
> > >
> > > A very dirty summary, but with a
nice picture representation of
what
> > > a whole bunch of superimposed
cycles actually looks like, can be
> > > seen at http://www.stockmarketcycles.com/technica.htm
>
>
>
>
>
>
> _____
>
> YAHOO! GROUPS LINKS
>
>
>
> * Visit your
group "equismetastock
> <http://groups.yahoo.com/group/equismetastock>
" on the web.
>
> * To
unsubscribe from this group, send an email to:
>
equismetastock-unsubscribe@xxxxxxxxxxxxxxx
>
<mailto:equismetastock-unsubscribe@xxxxxxxxxxxxxxx?subject=Unsubscribe>
>
> * Your use of
Yahoo! Groups is subject to the Yahoo!
> <http://docs.yahoo.com/info/terms/>
Terms of Service.
>
>
>
> _____
YAHOO! GROUPS LINKS