Hello ! This is a long post... hum.. let's go so. : )
First, ok it is moved and merged with this thread... and will try to classify smoothing method so we can start discussion from this. After take some times on the net to see some other smoothing technics i arrive to this point. There is : - parametric smoothing - non-parametric smoothing Parametric is good if you need to catch the model (for predicting or compute some statistics based on the model) or if you know some parameters of the signal (wich is not the case for stocks i think) Non-parametric is good (an easiest!) if you need to filter noise So for now i will focus on non-parametric method. There is : - moving average (all the MA, TEMA, T3, KAMA...), that we use. There are some linear (MA...), some non linear (aaptive T3, KAMA,...). - local fit smoothing : savitzky-golay, spline smoothing, binomial smoothing (seems like Gaussian smoothing say some people, is it right ?). - spectral method: time -> freq by wavelet, fft. Keep only dominant freq. and after freq -> time MA we got alot... : ) Spectral method : ok we got some (for exemple the trigfit i posted recently can be used for that...). I will focus now on local fit smoothing, because this area seems empty here... and i didn't know a lot about those technics. Three question part : 1- I will maybe try to make some savitzky-golay filter (code is in Numerical Recipes), but someone here try this ? Does it deserve to try to code it ? >From what i have read, from S-G filter, you can compute the coeffs of line and apply the filter to any data after. 2- Spline smoothing, don't do create filter coefs ? Is it same than B-Splines (use for interpolation and code is in Numerical Recipes too). You need to compute fitting for each data series ? Spline seems to need sometines some knots to anchor it (am i wrong ?). Is it a way to anchor the splines by creatin knots to the high volatile or big volume datas ? As you can see, the aera is unknow for me and i think maybe there is some things do do here ? 3- binomial smoothing, i don't find many explication on the web. Just this : http://www.thermo.com/com/cda/resources/resources_detail/1,2166,13425,00.html And this say binomial smoothing is better than savitzky-golay filter : http://www.wavemetrics.com/products/IGORPro/dataanalysis/signalprocessing/smoothing.htm So i don't have to bother to code S-G filter and use this more simple algo for binomial smoothing ? I suspect binomial smoothing to be a savitzky-golay filter, but with first order polynoms ? And i suspect to that smoothing splines to be near same than savitzky-golay filter, as booth fit polynoms ? But splines seems more powerfull. Savitzky-Golay are old (1967...) something new an dmore powerfull should be arround : ) Good read here about all this : http://vera.esalq.usp.br/~jorge/clarice/brazil_projector.pdf http://hug.phys.huji.ac.il/PHYS_HUG/MAABADA/Mabada_b/curve%20fitting%20in%20Matlab.pdf Sorry for so much questions. Maybe someone here can highlight this and say what has been done here with AB, what is usefull or better base on experience or reading, etc ... Just last things : good point for local fit methods over MA is that they keep maxima/minima and the original width of the cycle of the data. And if they use future data for fitting (ie: [-3;3] window instead of [-3;0] for S-G filter for exemple), there is zero lag ! But the last 3 filtered data points are not consistent as they can move depending on the data to come later. Like T3, TEMA, etc... (several EMA pass) we can make several pass of local fit to have more smoothing with only short lag added and some smoothing improvment. Cheers, Mich ----- Original Message ----- From: Rakesh Sahgal To: [email protected] Sent: Friday, December 08, 2006 2:58 AM Subject: Re: [amibroker] Jurik Moving Average JMA (working for me) I am aware that this forum is dedicated to solving problems realted to usage of AB and not discussing trading methods or philosophies. Instead of seeking out some magic moving average formula, I will humbly submit that maybe it will be more instructive if this thread moves to the AB-TS forum and focus is shifted a wee bit to effective usage of movng averages. Best Regards R _________________________________________________________________ Découvrez le blog Eragon sur Windows Live Spaces! http://eragon-heroic-fantasy.spaces.live.com/
