Machine learning in trading: theory, models, practice and algo-trading - page 2308
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
In my opinion, all traders who understand mathematics at least once tried to apply spectrum theory to the markets. I (like many others) did not find anything there and I do not really understand how anything can be found there.Still, we all make mistakes sometimes and it is quite possible that someone found something and is silent for understandable reasons).
There are still others who are threatening to find something) On them all our hope).
In the beginning of my acquaintance with Forex I tried to apply the classic Fourier method to a price series - to find a spectrum in the interval, select the basic harmonics, extrapolate them, find pivot points using their sum and trade. Naturally, the result is zero. Then I had the caterpillar and window Fourier followed by the wavelets, which were also window Fourier but with a specific window. Then I understood that spectral methods were not applicable to series with random genesis.
The first thing I did in the beginning of my acquaintance with Forex was try to apply the classical Fourier to a price series - find a spectrum on the interval, select the basic harmonics, extrapolate them, determine pivot points by their sum and trade like that. Naturally, the result is zero. Then I had the caterpillar and window Fourier followed by the wavelets, which were also window Fourier but with a specific window. Then came the realization that spectral methods are not applicable to series with random genesis.
Obviously, forex is the business of those for whom the opinion of others (which does not coincide with their own) is of little importance. Therefore, attempts to apply Fourier will always continue because of the obvious oscillatory nature of prices.
To which functions does the Fourier transform not apply?https://studfile.net/preview/1489143/page:2/
The main problem with tsosniks here is that they try to extrapolate a bouquet of sinusoids into the future. It doesn't work that way, because the overall error quickly grows
It's the trading logic, not the sinusoids, that needs to be validated on new data. When the cycles are found and the correct logic is written for them.
Fourier should be used only for exploratory analysis, to determine parameters of TS. And then learn either MO with regularization or on the basis of statistical observations.What functions the Fourier transform does not apply to?https://studfile.net/preview/1489143/page:2/
Rather, we are talking about the discrete Fourier transform, which is defined for any finite sequence of numbers.
Rather, we are talking about the discrete Fourier transform, which is defined for any finite sequence of numbers.
It doesn't matter
the important thing is that the Fourier transform MUST make sense for periodic functions or for functions with a finite number of extrema.
CR these are piecewise repeatable functions, what we are looking for are patterns or whatever you want to call them
and all that can be found using DSP is just detecting these patterns on the history - it's a very complicated task - it has been solved millions of times using different methods, after detecting a pattern, the price goes ... as usual to the right
SZY: Neat (Neuroevolution of augmenting topologies) is an interesting method, something between GA/GP and NS, it is a pity that the materials are in Bourgeois - not convenient
I made a fractal Brownian motion, but I couldn't find any way to do it other than PF. Bottom line. Hearst is related to the color of the noise. By changing the slope of frequencies you can change the herst.
Now for the fun part, follow the hands. Persistence is trendiness and antipersistence is flatness. You can pick your own profitable strategy under them. So if I can turn SB into a persistent series, then I can predict/earn on random?
1) I haven't seen any article yet that shows a meaningful difference between Hearst for real prices and 0.5 (value for SB)
2) Persistence != trending. SB with non-zero drift is obviously trending, but the increments are independent.
The main problem with tsosniks here is that they try to extrapolate a bouquet of sinusoids into the future. It doesn't work that way, because the overall error quickly grows
It's the trading logic, not the sinusoids, that needs to be validated on new data. When cycles are found and the right logic is written for them.
Fourier should be used only for exploratory analysis, to determine TS parameters. And then to learn either MO with regularization or on the basis of statistical observationsThe main problem lies in different conditions and objectives. The classical signal processing, the prior knowledge of the signal presence, its nature, and the task to find (clear) the signal or show its absence.
On forex, there is no knowledge about the nature of the signal, we look for stationarity, which is not stationary by nature, so the result is possible only on history, and in the future is possible only if the external conditions on forex are unchanged, which is extremely rare)))
Not only Fourier should be used)))))
Now let's forget about radio signals and their processing in this thread.
Off topic...
The main problem is in different conditions and purposes. Classical signal processing, the known knowledge of the presence of the signal, its nature, and the task to find (clear) the signal, or show that it does not exist.
On forex, there is no knowledge about the nature of the signal, we look for stationarity, which is not stationary by nature, so the result is possible only on the history, and in the future is possible only if the external conditions on forex are unchanged, which is extremely rare)))
It's not just Fourier that needs to be used)))))
cycles do not change by 5 years on fore, only a blind man would not see it. This is more than enough to write a simple test TS
Fourier is interesting to look for them on lower timeframes, maybe on ticks.
The problem, as always, is laziness.