Fourier connoisseurs... - page 8
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
If you don't mind, you can see the code that generates the continuation... you can drop it in your email...
It's kind of standard, I fill the indicator buffer with a certain index shift, for example, equal to the period of the Fourier curve, and then shift the indicator itself visually by the same value using SetIndexShift(0,Period) method;
I will post the code in the base later, when I will get it in order.
I beg to differ, let's assume we are at the end of the movement and in 10 points the trend will change,
I think we should not jump on the bandwagon, especially because the reliability of these 10 points is under question.
I have often noticed that the first 10 points are not true, but the nearest real quotes are equal to the forecasted ones.
Here the question flows smoothly into "Fourier or last point effect", and on this question it seems to me that the effect
is caused by another effect. Try to set a straight line of the form y = k*x + c, and then extrapolate with Fourier,
and instead of an upward straight line we get a downward curve. I would call it the incomplete wave effect.
I.e. if the wave does not fit in the measurement section, correct forecasting using Fourier method is impossible.
Both forward and long period harmonics are subject to this effect.
Therefore, in my indicators, I have made decomposition not relatively price=0, but relatively detrending lines following the ANG3110 example. Linear regression and Fourier interpolation of a larger period are used as detrending lines.
And I use Fourier interpolation if I manage to detect cyclicity over a longer period, otherwise I use LR. In this case the "incomplete wave effect" disappears.
I use Fourier interpolation if I can detect cyclicality over a longer period...
And how do you detect cyclicality? What method, what criteria do you use?
I do a Fourier extrapolation (those are clever words :) and look at the correlation between the result and prices. If the correlation is significant, it means that there is a pronounced cyclicality. Although there are probably better methods, I'm going to build a spectrum analyzer for MT
I do a Fourier extrapolation (those are clever words :) and look at the correlation between the result and prices. If the correlation is significant, it means that there is a pronounced cyclicality. Although, there are much better methods, I'm going to build spectral analyzer for MT
I understand the method, thanks for the answer, I think it has a place. And concerning words extrapolation, interpolation, approximation, correlation so the theme so who is not interesting let in the laggards chat, and who does not know so wikipedia is.
where the series is of a random nature no spectral m.Fourier,
The spectral extrapolation function cannot be discussed-this is wrong!
And you can and should calculate
spectral power density (SPM), i.e. the variance, the emissions,
whose amplitudes are distributed over the frequencies.
as a simple forecasting aid, I would recommend
A.A. Minko "Forecasting in Business with Excel",
and on Fourier analysis here, a classic of the genre, so to speak:
Jenkins, G., Watts, D. "Spectral Analysis and its Applications".
http://lib.mexmat.ru/books/853
http://www.newlibrary.ru/author/dzhenkins_g___vatts_d_.html
or here
S.L. Marple's "Digital Spectral Analysis".
http://prodav.exponenta.ru/read/info02.htm
If the above links do not fit, there are plenty of them to search for.
where the series is of a random nature no spectral m.Fourier,
We cannot talk about a spectral extrapolation function - this is incorrect!
And you can and should calculate
spectral power density (SPM), i.e. the variance, the emissions,
whose amplitudes are distributed over the frequencies.
you can do both
power estimation and series decomposition into functions have advantages and disadvantages
it is possible to do both
power estimation and series decomposition into functions have their advantages and disadvantages
Of course, but for forecasting it's very dangerous - non-linear methods
work fine inside the fitting interval, but outside it, when extrapolated,
the behavior becomes very insidious, so to speak.
It's very difficult to maintain such a predictive tool - because of
it is very insidious and also, as I said before,
incorrect.
Although, if you think about it, it's justified to apply m.Fourier to
moving average(MA), a fairly smooth MA, then yes :)
plus some regression straight line:
Fourier synthesis + regression polynomial (linear).
That's a pretty good combination.