|
Hi, Clyde
The real situation with
the application of spectral analysis to price time series looks this
way:
This is a spectrogram (Fourier transform,
more exactly spectral dencity) for DOW calculated using data for the
last 50 years:
I used bar metric here, so it reveals the
cycles in bars (it can be done in time metric as well). The red diagram is
Fourier, the black one is non lag moving average for Fourier.
It really shows some maximums around 26,
53 and 91 bars.
Here is the problem: beside these 3
cycles, there are a lot of other cycles that can play an important
part here and be relevant to the forecasting. The energy concentrated in
these 3 cycles is very small (take the energy as an area under this red curve).
In other words it is practically impossible to reveal some cycles here, the
situation is pretty close to so called "white noise" (when all cycles
are presented equally).
In reality using optimization methods I
can easily fit these cycles to the real price.
Like here I make a superposition of 26, 54 and 92 bars cycles
(with 2 overtones) optimized for Dow:
But when we conduct a backtesting
procedure for these cycles, the result is close to
zero.
The alternative approach (technology we
developed 3 years ago) is multiframe spectrum: it is less noisy, it is mostly
concentrated on latest cycles (so we need to recalculate it when the new
prce history comes).
Here it is for the latest Dow
data:
This is very important. The cycles
live their own lives, they change their phase and
period.
You can see it on the wavelet diagram (I
have marked there 90 days cycle, you see that it has started to play in
2007, before 2006 it practically did not work, besides its period started
to decrease in 2007) :
If the cycles would be permanent (as they
were considered at Fourier time, 18th century), we would see there parallel
straight lines. Now, knowing about Chaos theory, and some other things, we see
maybe not so beautiful pictures though more close to stock market
reality.
Now we try to apply another methods
(nonlinear filters) to cycles analysis as it is not a good practice to use one
procedure to reveal cycles and another to optimize them.
Best regards.
Sergey.
----- Original Message -----
Sent: Sunday, February 17, 2008 10:44
PM
Subject: Re: [RT] Next Week
In the attached chart we can see that the Fourier analysis
shows periods
of 26, 53, and 91 bars to be significant.
The attached pictures show
what is in the future for each
of these periods based on most recent 2
cycles of data.
The 53 period appears to be most significant right
now
and may be a clue to current price behavior.
Clyde
- - -
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Clyde
Lee phone: 713.783.9540
SYTECH Corporation
7910 Westglen, Suite
105
Houston, TX 77063 fax: 713.783.1092
WebSite:
www.theswingmachine.com
- - - - - - - - - - - - - - - - - - - - - - -
- - - - - - - - - - - - - -
----- Original Message -----
From:
"BobsKC" <bobskc@xxxxxxxxxnet>
To:
<realtraders@yahoogroups.com>
Sent:
Sunday, February 17, 2008 03:02 PM
Subject: [RT] Next Week
> So,
what do you folks think? I can make equally compelling arguments
> up
or down .. long term or short term.. near the end of the down turn
> or
just the beginning .. the bad news is factored in or it is ignored
> ..
you name your position and I'll argue it with you or against
> you.
Friday was pretty impressive considering the news.. I feel most
> of
the public sentiment is fueled by the press, (who have nearly zero
>
understanding of what they are reporting). I doubt if most members
> of
the press could tell you what makes for a recession but they love
> to
scare people. Anyway, I'm not usually at a loss for an opinion
> but
right at this point in time, I'm confused since I can justify
> either
direction.
>
> Have a great extended weekend all
>
>
Bob
>
>
>
>
> Yahoo! Groups Links
>
>
>
__._,_.___
![]()
__,_._,___
| 
