Machine learning in trading: theory, models, practice and algo-trading - page 2081
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
Well, does anyone else have any ideas how to do this, a grail in fact, by the way. With this technology, you can make any TS smart and adaptive.
It's adaptive filtering, Wiener filter, Kalman filter.
The dumbest thing is to take Fourier or wavelets, find the maximum frequency, and use it to choose the period of the mask.
With Fourier, you can zero out the uninteresting frequencies in the spectrum, the inverse transformation will give the filter
This is adaptive filtering, a Wiener filter, a Kalman filter.
It is not adaptive filtering, it is finding the best control curve.
In adaptive filtering we do not know the future, we optimally/adaptively adjust to the current process
This is the problem of synthesis of the best function, we can look into the future, we need to find the best control knowing the future
The dumbest one is to take Fourier or wavelets, find the maximal frequency and use it to choose the cycle time.
With Fourier, you can null the uninteresting frequencies in the spectrum, inverse transform will give the filter
Fourier is an option, but not as a dancing with frequencies, but as a tool for synthesizing an unknown function... adding different harmonics we will get different functions, maybe it will be so coincident that we will synthesize what we need, but the overshoot will probably be exorbitant.
But maybe there's an easier way.
This is not adaptive filtering, this is finding the best control curve.
In adaptive filtering we do not know the future, we optimally/adaptively adjust to the current process
This is the task of synthesizing the best function, we can look into the future, we need to find the best control with knowledge of the future
Fourier is an option, but not as a dancing with frequencies, but as a tool for synthesizing an unknown function... adding different harmonics we will get different functions, maybe we will get the right one, but the overshoot is probably too high.
But maybe there's an easier way.
In Kalman model of signal and interference is laid, it is assumed that we know the future, but in the market everything is complicated
The dumbest thing is to take Fourier or wavelets, find the maximum frequency, and use it to choose the period of the sweep.
Why do you think that the optimal control curve should correlate with significant frequencies in the speter?
Why should the DUT curve be periodic at all ?
Where are these thoughts coming from?
The Kalman model of signal and interference, we are supposed to know the future, but in the market it is complicated
Where is it supposed to be?
Anyway, don't answer that, it's not important, just trust me, it's not adaptive filtering.
You're not thinking what I'm thinking, so you're kind of looking at it from your own point of view.
Let me show you a simple example...
We have a trading system based on two scales with periods of 10 and 20, trading by crossovers, classic...
The scale is a low-pass filter, the period of the scale is the controlling parameter
In this case the controlling parameter is equal to the constants 10 and 20
the problem: on a given segment of the market to get DYNAMIC (not constants) control parameters of each sledgehammer, so they would be optimal in terms of profits
this is criterion №1
criterion no. 2 : received dynamic control parameters should be similar to a continuous function, not a chaotic dispersion of points.
How do you see the solution to such a problem?
And they will be continuous enough, if optimization is going by sliding window (0-100, 1-101....), but if you will optimize non-intersecting segments (0-10, 11-20....) then continuity will be gone, but it can be added by smoothing series of received coefficients, for example with a wizard.
The Kalman model of signal and interference is laid down, it is assumed that we know the future, but in the market everything is complicated
Here for illustration, I need to get what is ideal (red)
You can and should look into the future, all I need is to get the ideal control curve
Do you need this for your purpose? You move the Fourier window half a period forward (looking ahead) and get the actual parameters.
And they will be continuous enough, if optimization is going by sliding window (0-100, 1-101....), but if you will optimize non-intersecting segments (0-10, 11-20....) then continuity will be lost, but it can be added by smoothing series of received coefficients, for example by a wizard.
Well, yes, probably so I will try...
But it should not be smoothed - even the slightest smoothing distortion may affect "trade opened/not opened".
Do you need it for a purpose?
Yes
You move the Fourier window half a period forward (looking ahead) and you get the actual parameters.
Can you elaborate, for the dumb ones, how do I get the right periods for the mashups from the Fourier coefficients?