Have you had a look at Ehlers filter or the zerolag filter Ehlers is at the http://www.amibroker.com/library/detail.php?id=402 Zerolag is as follows for(; i < nSize; i++) { float a = 2/(range[i] + 1); result.array[i] = a * (data[i] + beta * (data[i] - data[i -3])) + ( 1- a)* result.array[i - 1]; } It seems that CT3 is the smoothest, havent tried Adaptive Laguerre yet One way to deal with the lag, instead of shifting the averages back, shift the predicted line back, so the lag is dealt with by the prediction.
Cheers Paul. ________________________________ From: [email protected] [mailto:[EMAIL PROTECTED] On Behalf Of Tom Tom Sent: Saturday, 2 December 2006 6:23 AM To: [email protected] Subject: Re: [amibroker-ts] Re: Moving Average... one more again : ) Yes true, so factor is 1. Before, i put the corretive factor in the loop because, it need to be corrected inside the loop to keep array "it" and array "array" same scale. But now it doesn't need at all factor héhé ! thx. I make comparaison with new formula, and it is now like very near a DEMA and not at all like a T3... oups ; ) Is it DEMA because there is two recursion on "it" array : it(i-1) and it(i-2) ? So this MA can be seens as two filter one FIR (act on the previous value of "array" and one IIR (act on the previous valus of the resulting MA "it"). The IIR is a DEMA. Diff. comes from the FIR part, am i right ? Ehler constructed this MA this way : - start from classic EMA - add a FIR part so impulse reponse is finite, so high frequency can be attenuated to 0 - take inverse of this LP filter ( 1-LP ) to obtain HP Filter - make HP Filter second order HP Filter by squaring its impulse response (signal pass two times in the same HP Filter) - make a final LP Filter from the HP Filter 2nd order ( LP_Final = 1 - HP_Filter_secondorder) So as this MA substract the high frequency (noise or Elher speak about cycles for those high freq.) of the original signal, and the low frequency are keep intact without any lag. That is why he says Zero Lag for low frequency (so long periods) .... so "Instatentaneaous Trendline". Keep only the trend without any lag by substacting cycle components. Héhé magic ehler ! But on charts seems like a classic DEMA... !? or TEMA. Not same than EMA because EMA keep a continious component which make it biased. You can clearly see this bias on the chart by plotting IT/DEMA/TEMA againt EMA. T3 filter a lot better noise, and can keep good amplitude by the slope factor... problem is it lag a lot more than IT/DEMA/TEMA for high periods. Maybe IT can be use as a good detrender. I will try to use it for AR prediction against the T3 i was using. For now we spoke about : - Instantaneous Trendline - DEMA/TEMA - EMA - T3 - Laguerre/Kautz Filter (i posted code to the library for the adaptive version : http://www.amibroker.com/library/detail.php?id=772 <http://www.amibroker.com/library/detail.php?id=772> ) Any new ones or advice about nice MA to used for finding trend (LPF) or cyclic (HPF) of a signal ? Cheers, Mich. ----- Original Message ----- From: Paul Ho To: [email protected] <mailto:amibroker-ts%40yahoogroups.com> Sent: Friday, December 01, 2006 7:01 AM Subject: RE: [amibroker-ts] Re: Moving Average... one more again : ) By the way shouldnt factor be factor = (alpha - alpha^2/4) + .5*alpha^2 - (alpha - .75*alpha^2) + 2*(1 - alpha) - (1 - alpha)^2; which is 1 anyway??? _____ From: [email protected] <mailto:amibroker-ts%40yahoogroups.com> [mailto:[email protected] <mailto:amibroker-ts%40yahoogroups.com> ] On Behalf Of Paul Ho Sent: Friday, 1 December 2006 11:17 AM To: [email protected] <mailto:amibroker-ts%40yahoogroups.com> Subject: RE: [amibroker-ts] Re: Moving Average... one more again : ) I am not sure if the correction is right, may be its better to use IT = IT /factor; after the loop. _____ From: amibroker-ts@ ts%40yahoogroups.com> yahoogroups.com [mailto:amibroker-ts@ ts%40yahoogroups.com> yahoogroups.com] On Behalf Of Tom Tom Sent: Friday, 1 December 2006 7:27 AM To: amibroker-ts@ ts%40yahoogroups.com> yahoogroups.com Subject: Re: [amibroker-ts] Re: Moving Average... one more again : ) Hi! I try the moving average. Thx ! Need some correction in the code to be exact scale (see end of message). I compared it to several MA. Different than Dema i found (ITrend seems to filter more). I find it to be nearer from T3. Try following settings : ITrend : 0.11 T3: Periods 6 and Slope 0.9 Near the same... ITrend is a Recursive FIR filter. (FIR part and FII part, numerator and denominator of impulse response). So if you say it is like Ingenious MA, maybe Ingenious MA is like T3... and any MA is the same as soon as you get good settings to match seems. That is what i think. All we do with all those linear MA is to change settings to move cut off frequencies and resonance or order of the filter. I find the best MA are those which are non-linear and take in account some another information to adapt their filtering (erreur from Yfiltered and Yoriginal, volume impact, ...). I try Laguerre Filter and even try to implement Kautz Filter (generic Laguerre filter... can have some impact one the phase because it treats complex signal). Kautz was to hard to code.... :). Laguerre filter, make differents frequency have not the same lag (lag depends of the frequency). More explaination on mesa elher website (there is code to compute it too, even an adaptive version) or look google for Laguerre or Kautz filter. Code for ITrend to be at good scale : //---------------------------------------------------------- SetBarsRequired(20000,20000); function ITrend(array, alpha) { // Choose One Of the initialization : // No pre-filter : it = array; // Pre-initialize its first value and after it converge to "array" value during main recursion (so act like no pre-filter after some times) //it = (array[2] - 2*array[1] + array[0])/4; //This initialization takes a long time to converge. // Pre-filter with classic triangular MA //it = (array + Ref(array, -1) * 2 + Ref(array, -2))/4; factor = (alpha - alpha^2/4) + 5*alpha^2 - (alpha - .75*alpha^2) + 2*(1 - alpha) - (1 - alpha)^2; for(i = 4; i < BarCount; i++) { it[i] = (alpha - alpha^2/4)*array[i] + 5*alpha^2*array[i-1] - (alpha - .75*alpha^2)*array[i-2] + 2*(1 - alpha)*it[i-1] - (1 - alpha)^2*it[i-2]; it[i] = it[i]/factor; } return it; } P = ParamField("Price field",-1); alpha = Param( "Alpha", 0.3, 0, 1, 0.01 ); data = ITrend(P, alpha); Plot( data, "ITrend", ParamColor( "Color ITrend", colorCycle ), ParamStyle("Style") ); //---------------------------------------------------------- Cheers, Mich ----- Original Message ----- From: Fred To: amibroker-ts@ ts%40yahoogroups.com> yahoogroups.com Sent: Thursday, November 30, 2006 6:07 PM Subject: [amibroker-ts] Re: Moving Average... one more again : ) Isn't this equivalent or roughly equivalent to a DEMA ... --- In amibroker-ts@ ts%40yahoogroups.com> yahoogroups.com, "Paul Ho" .> wrote: > >I use this one, looks close enough to what he is talking about. It is some >kind of FIR but it doesn't provide code. > >function ITrend(array, alpha) > >{ > >// initialize early array values and declare as array > >it = array; //(array + Ref(array, -1) * 2 + Ref(array, -2))/4; > >//it = (array[2] - 2*array[1] + array[0])/4; This initialization takes a >long time to converge. > >for(i = 2; i < BarCount; i++) > >{ > >it[i] = (alpha - alpha^2/4)*array[i] + > >.5*alpha^2*array[i-1] - > >(alpha - .75*alpha^2)*array[i-2] + > >2*(1 - alpha)*it[i-1] - > >(1 - alpha)^2*it[i-2]; > >} > >return it; > >} > >Its from Elhers' book. Alpha = 2/(period + 1); It has zero lag, I found >centering to be real bad for forecast as the averages finishes x bars ago >and AB copies the last known bar across these x bars, causing AR prediction >to flat line initially. > > > >_____ > >From: amibroker-ts@ ts%40yahoogroups.com> yahoogroups.com [mailto:amibroker- [EMAIL PROTECTED] roups.com> com] On >Behalf Of Tom Tom >Sent: Thursday, 30 November 2006 10:05 AM >To: amibroker-ts@ ts%40yahoogroups.com> yahoogroups.com >Subject: [amibroker-ts] Moving Average... one more again : ) > > > >Hello, > >I was just surfing and find this : >http://www.tradecis > <http://www.tradecis <http://www.tradecis <http://www.tradecision.com/product/special_techniques/ingeniousma.ht <http://www.tradecision.com/product/special_techniques/ingeniousma.ht> > ion.com/product/special_techniques/ingeniousma.ht> ion.com/product/special_techniques/ingeniousma.ht m> >ion.com/product/special_techniques/ingeniousma.htm > >Always competition between T3, Jurik, etc... and this one say it is the >better one (what a good advertisment! héhé). > >Does someone try it ? On what does it rely (noise filter adaptative based on > >statistics, use prediction so it can be a little shiffted backward, use >volume or volatility information or any other technics to filter noise, use >fitting method...etc ??). >It is consistent in time ?... . i mean, when a new data arrive it didn't >change old MA values (not like Fit or Predict. methods wich are not >consistent because some past values are computed from less old past value). > >Thanks. >Cheers, >Mich. > >__________________________________________________________ >Live.com: le nouveau moteur de recherche par Messenger! http://www.windowsl ><http://www.windowsl <http://www.windowsl <http://www.windowslive.fr/livecom/ <http://www.windowslive.fr/livecom/> > ive.fr/livecom/> ive.fr/livecom/> ive.fr/livecom/ > > > > > >[Non-text portions of this message have been removed] > __________________________________________________________ Live.com: le nouveau moteur de recherche par Messenger! http://www.windowsl <http://www.windowsl <http://www.windowslive.fr/livecom/ <http://www.windowslive.fr/livecom/> > ive.fr/livecom/> ive.fr/livecom/ [Non-text portions of this message have been removed] [Non-text portions of this message have been removed] __________________________________________________________ Windows Live Messenger sur i-mode : dialoguez avec vos amis depuis votre mobile comme sur PC ! http://mobile.live.fr/messenger/bouygues/ <http://mobile.live.fr/messenger/bouygues/> Send instant messages to your online friends http://au.messenger.yahoo.com
