![MQL5 - Language of trade strategies built-in the MetaTrader 5 client terminal](https://c.mql5.com/i/registerlandings/logo-2.png)
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
I agree. But neural networks have ways to avoid fitting to history, though there are complications there too. But with Fourier, there is almost no way to identify harmonics or harmonics that will be profitable in the future - that is the difficulty of applying Fourier to financial markets.
I.e. you can do these calculations, like decompose the data into harmonics, adjust amplitudes, phases, add up, only instead you can calculate coefficients to calculate the result as in FATL, SATL indicators - just multiplying prices by coefficients and adding up.
Agreed. But neural networks have ways to avoid fitting to history, though there are some difficulties there too. But with Fourier, there is almost no way to identify harmonics or harmonics that will be profitable in the future - that is the difficulty of applying Fourier in financial markets.
With any system, you should do a fit check, do a check test in the next section after optimisation. The same is true. With a network - train the network, test it. With any other system - optimise, check test beyond the optimisation period.
The way to determine the harmonic is to optimize in a tester.
With any system you should do a fit check, do a check test in the next section after optimisation. The same is true. With a network - train the network, test it. With any other system - optimise, check test beyond the optimisation period.
The way to determine the harmonic is to optimize in a tester.
Well, Fourier does not have such a method of determination - it is either guessing by coffee grounds or a finger in the sky, because everything rests in the choice, namely in the choice of harmonics. That's why Fourier has not found its application in the finnets.
Optimisation in the tester for Fourier is actually a choice of harmonics by profit margins on past data, but not a choice for future profit margins.
There is no tester that checks harmonics for the fact of fitting. In MT4 - this is unrealistic to do.
The coefficients do not have to be constant... I can build a model, use it to calculate the transfer characteristic of the filter with the required characteristics and adaptive parameters, go (Fourier/Laplace -> Z-transform) to the discrete domain, convert the transfer characteristic to a difference equation and then... profit!)
OK, got it. This is already a classic DSP.
Fourier has no such way of defining it - it's either coffee-leaf guessing or a finger in the sky, because it all comes down to choice, namely in the choice of harmonics. That's why Fourier has not found its application in the finnets.
Optimization in a tester is a choice of harmonics by the magnitude of profit on past data, but not a choice to bring profit in the future.
There is no tester that checks harmonics for the fact of fitting. In MT4 - this is unrealistic to do.
Realistically. It is not fundamentally different from testing a neural network by training it. In the end we are interested in profit. By profit and see if the system makes a profit in the section following the optimization section.
Okay, got it. This is already a classic DSP.
There is one more nuance here. The larger the next segment after the optimization, the higher is the probability that found harmonics will quickly become obsolete (stop bringing profit) on future data. Reducing this section results in unreliability of the test.
There is another nuance here. The larger the next section after optimisation, the more likely it is that the harmonics found will quickly become obsolete (no longer yield profits) on future data. Reducing this segment - we get unreliability of verification.
No such problem with neural networks?
Yes, there is. But there are some patterns that can be noticed when training a network, and some training techniques that allow you to do without even a forward test. I don't know about Fourier and I haven't heard about it.