Machine learning in trading: theory, models, practice and algo-trading - page 2087

New comment
 
mytarmailS:

Karoch )) do not need anything, no frequency or phase, just feed one amplitude of one harmonic with the largest amplitude, just one sign

if you add something there is a lot of noise

and you get a very clear signal, another question is what this signal carries))

just noise is not audible

what you get is what you get

and the Fourier transform will highlight only what you want
 
Renat Akhtyamov:

you just can't hear the noise.

What you submit is what you get

and the Fourier transform will only highlight what you want

what you ask (if it was always the same want)))))

 
Renat Akhtyamov:

you just can't hear the noise.

just PSHNH :)

 
mytarmailS:

just PSHNH :)

Don't bullshit me.

he has a big amplitude and it's not clear what happened...

;))))

 
Valeriy Yastremskiy:

I think the task is more complicated. It is not important to take into account only news, the experience of traders proves it) It is necessary to watch the price change, but I would look for significant changes in price from the news in order to find what else influenced the price. Watch what news was around. And maybe mix it in with stock indexes or whatever else is in the number.

In addition, describing the news only in terms of significance is flawed. There is the expected value of the news, there is the real value, and there is the direction of the impact on the price.

I agree, reality is much more complicated than any of our models, but we can meaningfully cognize it only with the help of simple models.)

 
Aleksey Nikolayev:

I agree that reality is much more complex than any of our models, but we can only cognize it in a meaningful way with the help of simple models.)

It is possible (and necessary) to search for the synergy of signs using simple models, so that we can construct complex models with understanding. Nobody forbids to look at different single signs of influence on the price on similar tests, news, stock indexes, the Central Bank rate, third-party rates or something else and to group them in case a correlation is found.)

 
Aleksey Nikolayev:

I agree, reality is much more complex than any of our models, but we can only meaningfully understand it with the help of simple models) Such a dialectical paradox)

The reality is that we sell more expensive and buy from us cheaper.

the average trader works with averaging, whatever way you look at it

that's all

you have to be able to outrun both and, as a consequence, to stand out from the crowd

 
Valeriy Yastremskiy:

Synergy of signs can (and should) be searched for by simple models, so that we can build complex ones with understanding. Nobody forbids to look at different single signs of influence on the same tests, news, stock indexes, Central Bank rate, third-party rates, or anything else on the price and group them if a correlation is found.)

The FLEXIBLE is something that is CONNECTED by simple rules from the simple.)

 
Aleksey Nikolayev:

COMPLEX is that which is CONNECTED by simple rules from simple)

like spectral analysis, from simple harmonics, you can CONFIGURATE a function of any CONFIGURATION, an additive model...

If you think about it, the same Random Forest is also additive model, and in some sense spectral decomposition, though it's messy...

 
mytarmailS:

Thank you, really explained, I do not perceive formulas ((

I've read about filter types long ago, it's quite clear in principle.

I searched through my archives, found it, remembered it, figured it out, but it's still an interesting thing


The target - approximated series, I approximated "ssa", but Fourier is also possible, the main thing is to smooth it out nicely

Like this


Then in the sliding window

we do the Fourier transform, remove the harmonic with zero frequency, find the harmonic with the largest amplitude (as you said)

and give it to the MSUA to study

after the training we get an incomprehensible but very clear signal, you can clearly see that the harmonic

And here at the same data Forest did not find anything

I think how to distinguish such clean and useful signals from the spectrum ?? Very interesting thing this spectral analysis ... and filters on it...


By the way, a couple of pages ago they were saying that Forest can't interpolate data inside itself, but the grid can, here's a vivid example

Do Fourier based on the increments, you can even do without SSA.

For training purposes make an indicator with three sinusoids and overlay it with different indicators, you'll see what's what.

1...208020812082208320842085208620872088208920902091209220932094...3574
New comment