|
PureBytes Links
Trading Reference Links
|
From: tomy_frenchy <michel_b_g@xxxxxxxxxxx>
Hello,
After the technicals post about quotes predictions (hurst, polyfit) on
amibroker software, i was asking me if someone here explore different
prediction technics and if he could give us his feedback on accuracy of
prediction depending the forecast tools. It will be very nice.
1- Prediction with AR model
For now I am working on AR (Auto-Regressive model) prediction. I post code
here : http://www.amibroker.com/library/detail.php?id=757
You will find a jpg picture joined to this mail. The data before the
vertical bar are those used to compute the AR coefficient and behind the
vertical bar are those predicted (and out-of-sample). The data before the
verticale bar colored in blue are predicted too (because of CMA for
denoising who lag) but are in-sample. So i resume (sorry for my english
don't be affraid ; ) ) :
Green : computed CMA in-sample, Blue: predicted with AR model in-sample,
Red: predicted with AR model out-sample
Prevision are sometimes good (if data present enough cyclic pattern and
number of AR coeff is enough high to take in consideration the patterns).
Somtimes they are wrong too, as always héhé : ))
CMA is used here with what i call a false "T3 zerolag". So it denoise the
data and keep the averaging synchronized with the data.
2- Prediction with PolyFit / Hurst CMA
Thanks to Fred for this very nice code :
http://www.amibroker.com/library/detail.php?id=741
Thanks too to Andy for this code :
http://www.amibroker.com/library/detail.php?id=753
CMA is used here (manually (hurst de) or automaticaly (cycle highlighter))
to extract one (cycle highlighter) or differents (hurst de) MA with
different periods wich keeps synchronized with data, and prediction is made
with polyfit.
More information on the recent post named "Hurst Channel's".
3- Prediction with EMA/DEMA/TEMA (seems not interesting to go to higher
order prediction)
I posted code for DEMA prediction here :
http://finance.groups.yahoo.com/group/amibroker/message/102710
Principe: an EMA or DEMA or TEMA is computer. From this is computer a linear
regression or higher order regression. Data are predicted using the result
of this regression.
Forecasting is very poor and can indicate only the trend to come. It cannot
show future turning point and other variation, nor show pattern/cycle
repetition.
More information : http://www.duke.edu/~rnau/411avg.htm ,
http://espse.ed.psu.edu/edpsych/faculty/rhale/statistics/statlets/usermanual/sect6_3_3.htm
4- Prediction with different CMA automaticaly computed (CMA, downsampling
and interpolation, trigfit)
Post about his can be found here :
http://finance.groups.yahoo.com/group/amibroker/message/102530
For now it seems a very interresting approach. It could take automaticaly
pattern in consideration and is more consistent than AR prediction i think
wich sometimes can provides some divergent prediction (like polyfit with
high order).
Fred could you tell us more about this work and maybe post some code about
it ? Will be very nice.
5- Another prediction tehnics
We can find non-linear prediction (Kalman, Network prediction, etc...)...
seems hard to compute. Someone try this ?
For linear prediction : ARMA, ARIMA, Trigonometric Fit, ARCH, GARCH (for
volatility prediction)
Based on spectral analizis to extract cycle :
for several cycle : FFT (different version exist depending speed,
resolution), Density Spectrum Power,
Prony, Pisarenko, Wavelet
for the main cycle : MESA (Ehlers Dominant Cycle Period), CMA
Pisarenko method seems to offer better resolution from FFT and less
constrain, but suffer that is is more difficult for computing it.
So can you provide us some information on your work on prediction with
amibroker. Tips, code, experience, trading result in applying prediction and
anything who can help to go forward in this topic is welcome.
Cheers,
Mich.
_________________________________________________________________
Retrouvez tout en un clin d'?il avec Windows Desktop Search !
http://desktop.msn.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
|