Fast Fourier Transform - Cycle Extraction - page 104

 
mladen:
Tsar

It does not need any change (neither the mq4 file nor the dll file)

Here is how it looks on my chart without any change in code at all

In my ECN 5 digits terminal need use Start/Resumed debuugging and that's Indie work again.

(I didn't installed before) No Problem with my 4 digits terminal. Btw.. found variable 'tPeriod' not used (Line 222) I wish not make a Problem

 

great indicator

 
Tsar:
In my ECN 5 digits terminal need use Start/Resumed debuugging and that's Indie work again. (I didn't installed before) No Problem with my 4 digits terminal. Btw.. found variable 'tPeriod' not used (Line 222) I wish not make a Problem

Tsar

That 'tPeriod' not used (Line 222)" is completely benign compiler warning. It des not affect the indicator work at all

 
Tsar:
In my ECN 5 digits terminal need use Start/Resumed debuugging and that's Indie work again. (I didn't installed before) No Problem with my 4 digits terminal. Btw.. found variable 'tPeriod' not used (Line 222) I wish not make a Problem

It works OK even with that warning

 

Eurika... After search more One Week, finally I found it

Compiled with MT4 Build 830 didn't found an Error or Warning.

The Problem solved in my ECN Terminal 5 digits.

 

unfortunately FFT it turned out to overfit so to be useless for trading

https://www.youtube.com/watch?v=D8e3FcySitY&feature=youtu.be

Krzysztof

 
fajst_k:
unfortunately FFT it turned out to overfit so to be useless for trading

https://www.youtube.com/watch?v=D8e3FcySitY&feature=youtu.be

Krzysztof

Using things like Fourier analysis on a non-stationary data is never going to give results that are usually expected - that is why it is usually used for estimation.

And the question of how many frequencies are there in financial time series remains highly arguable (there is no exact mathematical way to determine it exactly). Goertzel algorithm usually gives some 15-20 significant frequencies which would mean that the Fourier transform still fits in the expectations even for financial time series

 

My mail was to show how easy is to fool FFT to overfit i.e. search a cycles in noise not in signal. For financial markets where S/N ratio is very low so it will be a case. Unfortunately it is a case in the most of backtests or even designing of the indicators due to selection bias. See this. BTW cycles in the example were stationary.

3rd generation NN, deep learning, deep belief nets and Restricted Boltzmann Machines - Page 24

Krzysztof

 
fajst_k:
My mail was to show how easy is to fool FFT to overfit i.e. search a cycles in noise not in signal. For financial markets where S/N ratio is very low so it will be a case. Unfortunately it is a case in the most of backtests or even designing of the indicators due to selection bias. See this. BTW cycles in the example were stationary.

3rd generation NN, deep learning, deep belief nets and Restricted Boltzmann Machines - Page 24

Krzysztof

Well I agree on one thing : overfiting is one of the man causes of wrong trading decisions (and in most TA analysis in general).

 
mladen:
Was thinking what would be a logical way to calculate bands for Fourier extrapolation of some indicator. This would be one possible way : it uses standard deviation of indicator values in order to draw bands. It seems to be giving logical results with 1 standard deviation used for bands width and is also very easy to adapt it to show Fourier extrapolation of any indicator

As an example : this one is a Fourier extrapolation of stochastic

PS: if sombody wants to avoid current bar calculation, simply set the LastBar to some value other than 0 (1 would be a first closed bar, and so on)

Hello mladen,

the Fourier extrapolation dll is not allowed anymore in the actual Metatrader clients. Could you check and update this ?

thanx and all the best

Reason: