Seeing into the future...with Cycles.

August 13th, 2009, 10:04 AM

Quoting

wh are you working with the files from bearcave.com? I tried your indicator but doesn't look like its drawing correctly here.

No, i worked with the nonlinear workbook (g****e,rapids***e). the code is at the time limeted. with bearcave I do not get good results

some things must be changed. now i count only 54 bars back and not the full series. the other question is how i interpret the wavelet correctly. i mean the haar wavelet use only 1,0,-1 for this. For more infos i used wiki, in this case my math skills are sleeping well.

Code

                            
            int t,tt;
            double[] x = new double[n];
               for(t=0,tt=n;t<x.Length;t++)
            {
                x[t] = Close[tt];            // time series
                tt--;
            }

Code

                            
            for(t=0;t<dd.Length;t++) { sum += Math.Log(dd1[t]/dd[t]); }
            double lambda = sum/((double) (dd.Length));
            
            //Print(CurrentBar +" lambda = " + lambda);
            
            if(lambda > 0.0 && lambda < 5) P.Set(1);
            else if(lambda < -0.0 && lambda > -5) N.Set(-1);
            else Z.Set(0);

wh

August 13th, 2009, 10:10 AM

i will make some changes, the next days.

August 28th, 2009, 03:40 AM

wh, are you still pursuing wavelets?

They are a bit hard to understand at first but it is a very powerful methodology for time series analysis, worth spending some time to learn IMHO.

BTW with Haar transform you don't have to worry about boundary conditions and edge effects that I mentioned previously. The only thing you must worry about in that case is that you use an undecimated version of the transform (no downsampling).

StationaryWT, MODWT, Translation Invariant WT, A'Trous algorithm are all different names for the undecimated transform.

You can get very good denoising result using only the Haar transform provided that you use good thresholding method. Check this research paper which gives simple version of the algorithm: https://www.unige.ch/fapse/mad/renaud/papers/KalmanWavelet.pdf

September 2nd, 2009, 03:57 AM

sefstrat

wh, are you still pursuing wavelets?

They are a bit hard to understand at first but it is a very powerful methodology for time series analysis, worth spending some time to learn IMHO.

BTW with Haar transform you don't have to worry about boundary conditions and edge effects that I mentioned previously. The only thing you must worry about in that case is that you use an undecimated version of the transform (no downsampling).

StationaryWT, MODWT, Translation Invariant WT, A'Trous algorithm are all different names for the undecimated transform.

You can get very good denoising result using only the Haar transform provided that you use good thresholding method. Check this research paper which gives simple version of the algorithm: https://www.unige.ch/fapse/mad/renaud/papers/KalmanWavelet.pdf

i look around on this wavelet, but it is not mandatory for my trading. i like it quite simple and wavelets more then difficult to identify.

but it is interesting for me: to create in the future a hybrid of ann + wnn. so i must learn a lot.

wh

https://users.cs.fiu.edu/%7Echens/PDF/ICTAI02_WNN.pdf

September 2nd, 2009, 07:53 AM

wh

i look around on this wavelet, but it is not mandatory for my trading. i like it quite simple and wavelets more then difficult to identify.

but it is interesting for me: to create in the future a hybrid of ann + wnn. so i must learn a lot.

wh

https://users.cs.fiu.edu/%7Echens/PDF/ICTAI02_WNN.pdf

True, wavelets are not that useful for discretionary trading, it is very powerful for algorithmic trading due to its power of feature extraction which makes pattern recognition much easier.

I skimmed through the research paper you linked above, only problem is their method can't be used in real time because they use future data in their calculations. Here is another similar paper which uses wavelets, fuzzy logic and neural network together for forecasting exchange rates.

This one I have implemented myself and can verify that the approach works quite well =)

September 2nd, 2009, 03:48 PM

thanks my inputs based all on this books

one from the good old algol area ...

and the other is the nonlinear workbook

September 3rd, 2009, 05:17 AM

some interesting research papers:

https://mpra.ub.uni-muenchen.de/2482/1/MPRA_paper_2482.pdf

July 9th, 2010, 07:11 PM

Thanks! Do you have the toolkit in zip form? Thanks

August 26th, 2011, 08:49 AM

A few years old but a very interesting thread. I'm posting here to get notifications for updates : )

BTW I have a phd in signal processing, but relatively new to algo trading. Happy to collaborate with anyone on developing / researching these algos

September 20th, 2011, 04:08 PM

Hi, NMTrader,

I like your cycle analysis. are you planning to post more pictures of cycles?

Seeing into the future...with Cycles.

Discussion in Traders Hideout

Seeing into the future...with Cycles.