Machine learning in trading: theory, models, practice and algo-trading - page 2675

 
Aleksey Nikolayev #:

Here, for example, are the assumptions underlying the Kalman filter. The possibility of separation into signal and noise is assumed from the beginning, not proved. For some physical object these are quite obvious assumptions, but not for prices.

Of course, nothing prevents us from decomposing a price into components, but no component will be noise (a stationary independent process). Well, in my opinion, such decompositions should not be based on frequency separation, but on the separation of the low-amplitude component. Because of the presence of pulses, these are different things.

A community of traders in a sufficiently large number with approximately the same parameters can be compared to a medium similar to physical media. In Brownian motion we don't know and can't know the maximum and minimum speeds of objects, we can also determine the max min and average speeds only for a certain large percentage. It is possible to do the same for this number of traders in a static period. And it is quite possible that in such an environment of traders impulses will cause damped waves).

 
mytarmailS #:
Case study https://youtu.be/2JgoeuM7iVM

In this example, signal and noise objectively exist initially as separate things. This analogy does not apply to the market - there is only an initial single price, which we split into noise and signal only at our discretion.

 
There is no signal in the bazaar, there are inefficiencies. Buy boots from a grandmother, push them in exchange for a hat and sell it for more than boots.

These are not comparable things at all.
If you look at the history of profitable TS in certain periods, they all trade a specific inefficiency, which is then smoothed out.

And all old markets are generally efficient. This means that it is impossible to warm up other participants, as they are also not a finger on the pulse.
 
Valeriy Yastremskiy #:

The community of traders in a sufficiently large number with approximately the same parameters can be compared to a medium similar to physical media. In Brownian motion the maximum and minimum speeds of objects we don't know and can't, we also only for a certain large percentage can accurately determine the max min and average speeds. It is possible to do the same for this number of traders in a static period. And it is quite possible that in such an environment of traders impulses will cause damped waves).

The analogy is quite understandable, but it is not fully developed) Each "particle" of the market reflexes, tries to comprehend the market as a whole, etc. (like your reasoning, for example). This changes everything a lot and it is hardly possible to "catch" this "physics" by simple wave approaches.

 
Aleksey Nikolayev #:

In this example, signal and noise objectively exist initially as separate things.

Yes, that's right.

Aleksey Nikolayev #:

This analogy is inapplicable for the market - there is only an initial single price, which we split into noise and signal only at our discretion.

Yes, so what is the inapplicability?

The video example was to show that it is possible to separate noise from signal in real time mode (everyone decides what is noise and what is signal as I wrote above).

 
Maxim Dmitrievsky #:
There is no signal in the bazaar, there are inefficiencies. Buy boots from a grandmother, push them in exchange for a hat and sell it for more than the boots.

Not comparable things at all.

There are both, but they are really incomparable things.

Buy a product - somehow transform it (processed it, delivered it to the counter, etc.) - sell it - go back to 1 point and repeat all over again. What is this if not a cycle? Of course, there are a huge number of such cycles on various scales and due to their constant interference a clean signal is not obtained. Well, it turns out to be dirty, but you can still work with it.

 
mytarmailS #:

Yeah, that's right.

Yeah, so what's the unworkable part?

The video example was to show that it is possible to separate noise from signal in real time mode (everyone decides what is noise and what is signal as I wrote above).

In the example there is a microphone outside the earphone, which catches pure original noise. What can we have as their analogue (microphone and noise)? All the maths there is basically just determining how the noise is distorted as it passes through the headphones (and then this distorted noise is "subtracted" ) .

 
Aleksey Nikolayev #:

In the example, there is a microphone on the outside of the earpiece that picks up pure raw noise. What can we have as their analogue (microphone and noise)? All the maths there is essentially just determining how the noise is distorted as it passes through the headphones (and then this distorted noise is "subtracted" ) .

The video is an example of adaptability, not a direct guide to action

The video shows that it is possible to deal with a changing environment, and that conventional non-adaptive filters cannot cope with a changing environment....

1. The market is a changing environment

2. indicators, neurons - non-adaptive attributes/filters
 
vladavd #:

There's both, but they're really not comparable things.

You buy a product - somehow transform it (recycle it, deliver it to the counter, etc.) - sell it - go back to 1 point and repeat all over again. What is this if not a cycle? Of course, there are a huge number of such cycles on various scales and due to their constant interference a clean signal is not obtained. Well, we get a dirty one, but we can still work with it.

Well, it is a trend, but not a signal ). A signal is when there is some ordered information that can be interpreted, right?

Can wheel marks on the road be considered a signal? I guess if someone
left them deliberately, yes. But what's the point of extrapolating it.
Or you could start analysing and extrapolating a forest trail or a powder track
 
Maxim Dmitrievsky #:
Well, it's a trend, but it's not a signal )

Well, let's subtract the trend, normalise the amplitudes - we will still have a sine wave(s).

Using the same conditional commodity as an example: we stock the warehouses for production - the price rises, we stocked up - it stopped growing or started falling. After a while, having sold out stocks, we return to the market and sell again - the price rises again. It is clear that in reality everything is much more complicated, but the basic model is exactly the same.


Maxim Dmitrievsky#:
A signal is when there is some ordered information that can be interpreted, right?
.

What is easier to interpret than a sine wave? Sell at the top, buy at the bottom.

Maxim Dmitrievsky#:
Can wheel marks on the road be considered a signal? Probably, if someone
left them deliberately, yes. But what is the point of extrapolating it.
Or you could start analysing and extrapolating a forest path or a powder track.

There is no need to consider trails as a signal, because we have a road in front of our eyes, so there is nothing to predict - just look ahead. There is no road on the market, there are only vague ideas of what it will be like.