Fast Fourier Transform - Cycle Extraction - page 31
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
That's wicked man ,wicked
I am sure that Pava did not think bad
Winograd in Russian means grapeyard and as far as I see he added some grapes to that chart
I am sure that Pava did not think bad Winograd in Russian means grapeyard and as far as I see he added some grapes to that chart
''Wicked'' means '' excellent,brilliant'' in Ali G's vocabulary
My bad. Sorry
''Wicked'' means '' excellent,brilliant'' in Ali G's vocabulary
Spectral Zero Lag Indicator
Anyone has heard of Spectral Zero Lag Indicator ? Anyone can do it ?
Here it is. Thanks : Spectral Zero Lag Indicator: Introduction to the Spectral Zero Lag indicator for Meta Trader 4
If you take a look at the comparison to traditional averages you will notice that it goes ahead of the price in the rounded part. That is a classical SSA calculation and they should have told that it recalculates (but that part about the recalculation the "forgot" to mention)
Anyone has heard of Spectral Zero Lag Indicator ? Anyone can do it ? Here it is. Thanks : Spectral Zero Lag Indicator: Introduction to the Spectral Zero Lag indicator for Meta Trader 4
If you take a look at the comparison to traditional averages you will notice that it goes ahead of the price in the rounded part. That is a classical SSA calculation and they should have told that it recalculates (but that part about the recalculation the "forgot" to mention)
Thanks a lot techmac
. I thought there was some hidden-non told information about it.
Looking at the site, it states that the indicator is using wavelets to calculate the length of the cycle but there is little information there. If it looks to good to be true.....
Wavelets come up once in while and rather than read my two cents, here is a cut and paste from Mark Jurik in his paper "Evolution of moving averages".
"Fractional Brownian Motion (fBm) has long been considered a plausible model for financial asset markets. A fractal
structure of the market, indicating the presence of correlations across time, hints at the possibility of some predictability.
Recent advances in time/frequency localized transforms by the applied mathematics and electrical engineering communities
provide us with new methods for the analysis of this type of process. In fact, it has been proven by Wornell that the wavelet transform with a Daubechies basis is an optimal transform for fBm processes. ...
"With this result we consider using the wavelet transform/multiresolution decomposition to analyze financial time series.
Specifically, the discrete wavelet transform can be used to decompose a signal into several scales, while maintaining time
localization of events in each scale. In terms of financial time series, we can conceptually think of each of these scales as the dynamics associated with information and traders of each investment time horizon. For instance, long term traders, such as institutional investors, basing their trades on long term information, form the low-frequency component of the market. Once we have extracted out these scales we can view each as a stationary time series, which can be modeled, analyzed and predicted individually, either independently, or in conjunction with other scales and data that is relevant to that scale."
Right...so what does it really look like? I created a script in rapidminer which will perform a maximum overlap discreet wavelet transform using the Daubechies wavelet. The definitive text on this is "Wavelet Methods for time series analysis" by Donald B. Percival and Andrew T. Walden. Not light reading but there are other texts particularly "An intuitive guide to wavelets for economists" by Patrick Crowley which will be a useful introduction.
Here is a wavelet decomposition of the last 128 days of EURUSD- Using dim 5-7 as an example. Separating the cycle from the noise is still problematic with wavelets and there is nothing to suggest that the cycle will persist in the future. From the plot, we can see some cyclical activity which corresponds to a cycle of about 41 days (using only dim 5).
Here is MiniMe 2 cents
if you are looking to get an equation then Fourier is the best way to generate the equation from a curve , however if you only use Fourier analysis then you would face the following problem : Fourier assume cycling nature of the signal (sine, cosine ...).
So to decomposing the signal to its Fourier harmonics we need to define the cycle length , and for sure if you don’t know the cycle length and or don’t have a clear idea about your signal, exactly like what we have here in the forex market then we wouldn’t be able to know the cycle length .
Another problem with Fourier is that its not a good tool to detect fast and rapid changes in the signal, you would use many harmonics to simulate a rapid step response .
An alternative approach to solve the problem with the typical Fourier is Winding Fourier , where the signal is divided into smaller sections and each section is analyzed alone using Fourier, but again what’s the window size ?
Wavelet is known to be used in pattern detection and signal de-nosing , but you need to hard code your pattern. Unlike Fourier , Wavelet does not produce an equation , the results are mostly visual.
The power of wavelet is that you can think of wavelet as a rubber band , if you have a pattern you save it as a basis function and if you run the wavelet analysis you can find the scale and the time where this pattern have occurred. Therefore Wavelet is used for image compassion and pattern detection also unlike Fourier’s analysis , wavelet maintains the shape of the signal and doesn’t
The concept behind detection in wavelet is that we have a reference image or data and look for something similar to it somewhere else . Using wavelet we can do this in two ways : we can define the pattern as a basis function and use it for our reference detecting . Or we can store a two dimensional image and compared any other image with it till we have a match.
If we chose the fist option then there are some rules that needs to apply for the pattern before we can define it as a basis function , unfortunately not all patterns can follow these rules. The main simple rule as I remember is that the integration of the pattern should equal to zero.
If we chose option two then in this case we have to use 2 dimensional detection and this is the same technique used for face recognition or text recognition .
The concept is exactly the same as 1-D detection as we have a reference image and we use it for detection HOWEVER there is a problem, I’ll explain the problem by the following example :
if we asked two person to write the same aplanatic letter (A) you wont find two people that will give the same results , in such case we need to educate the software and teach it by using different images this is done using neural network (NN) or support vector machines (SVM).
In most finical studies Wavelet is used for de-nosing signals to get rid of high frequency noise. Not for pattern detection.
Its as if you pass the signal through a high pass and low pass filter and you have the ability to chose the threshold .
So bottom line Wavelet have its limitations , Fourier have its limitations.
The way around this problem is to combined (Fourier , wavelet/SVM and NN ) and divide the data into segments , one to educate your software , second for verifications and third is your prediction.
Regards,
MiniMe
In my previous post I meant Windowing Fourierand not Winding. ( spelling mistake )
What I meant from my previous post is that Fourier assumes a cyclic nature of the market , therefore you need to have an idea about the cycle length ; but if you have an idea about the cycle length then you don't need Fourier from the first place , just use the cycle lines on the chart to connect the turing points and place your trade, and in this care Fourier will be a nice added tool.
I am not bragging
but if you know the cycle then you can make > 500 pips per trade , check my screen shot this is the profit in pips/10 on the current open trades , and those are not fluke trades, I prove it to brokers that I can make more than 500 pip per trade. So if you can spot a cycle you can define where the market is heading for the next few months maybe even years .
I know it may sound like B.S but check the SP500 screen shot and check how I planed my trades , I plan to short SP500 around the mid of next year , again sound like B.S. but the truth is the market have cycles and you need to spend time to find them.
Regards,
MiniMe