|
PureBytes Links
Trading Reference Links
|
Hi !
I don't have lot of those "Subscript out of Range" on my computer exept if i
go to near the right border limit of the data (should have to use the SHIFT
feature of the Plot fonction i think). Same if i go to far near the begining
of the data.
I will try to correct those problem. It is just some constrain to add to the
choice of parameters.
For now i try to improve stability of coefficient calculation without to use
some complex criter to valide the AR model.
I find it is a lot more robust if you use the BIASED estimator (it is in
commentary // in the code source, you have to switch from non-biased to
biased one).
I try to make some specific ponderation and pre-modeling of the data to
highlight important information or dismiss useless ones.
Highligth: high volume = data to be more weighted, use the fractal (based on
elher work) wich is in fact a volatility indicator using High or low and
timeframe correlation (based on time frame N and N/2).
Purpose is to use Volume, High and Low, Timeframe information to accurate AR
prediction
Dismiss: i work in intraday so i dismiss the EOD gap. When big gap occurs,
it can ruins prediction because gap is too much strong compared to intraday
variation. This is to explore.
I try to make those new feature present in ARMA prediction (MA predicted by
iterative durbin algorytm based by simulating MA with long AR and minimising
ARlong estimation with least square).
For now it doesn't work at all ... : (( Need to fix the MA code...
So i join you just a picture of ARMA(50,0) (AR order 50, MA order 0).
Original data is computed from six sinus waves with differents amplitudes
and frequency. Linear trend is added and white noise too.
What is good is that is is more robust now (order 50... damn good). I find
it exhibits more small various pattern but miss long cycle.
A good way will be :
- use TrigFit to find some medium/long cycle
- substrat those to data
- use the new data to compute the AR model
So finaly, we got cycle and pattern together. ARMA and Trigonometric polynom
combined.
Pattern need sometimes many sinus to be represented (a square for exemple).
With this technics we have the long prediction by cycle and near prediction
by patterns (squared pattern which show fast trend change...).
We have to see if prediction is accurate or not (more than 50% héhé at
least).
Cheers,
Mich.
>Mich,
>
>Thanks for providing more more food for thought. What you've done looks
>interesting as far as it > goes, although I too get lots of "Subscript out
>of Range" errors.
>
>I agree with Fred's comments regarding Ehler...I wouldn't use *any* of this
>stuff if I didn't
>understand it 100%. Looks like your maths might be up to it too? Lucky you.
>
>Andy
_________________________________________________________________
Découvrez Windows Live Messenger : le futur de MSN Messenger !
www.windowslivemessenger.fr
Content-Description: "AVG certification"
No virus found in this incoming message.
Checked by AVG Free Edition.
Version: 7.1.409 / Virus Database: 268.14.11/542 - Release Date: 11/20/2006
|