Machine learning in trading: theory, models, practice and algo-trading - page 3706

[Deleted]  
There is an option to generate different market modes that work well on new data, pour them into codobase in the form of onnx. Then it can be used to optimise any TS on the selected market modes. It seems to me that this approach can work. You don't need anything from users except simple checking for the presence of a particular mode. The rest of the logic is their own.
 
Maxim Dmitrievsky #:
There is an option to generate different market regimes

What signs did Sorcerer use for clustering? Did he tell you?

[Deleted]  
Evgeni Gavrilovi #:

What signs did Sorcerer use for clustering? Did he tell you?

No.

 
Maxim Dmitrievsky #:
There is an option to generate different market modes that work well on new data, pour them into codobase in the form of onnx. Then it can be used to optimise any TS on the selected market modes. It seems to me that this approach can work. You don't need anything from users except simple checking for the presence of a particular mode. The rest of the logic is their own.
Imho, it will take off if the market has modes that (a) are clearly defined and (b) live long enough after they are defined.
 
Maxim Dmitrievsky #:

no

Is it already known which island he bought and that it is definitely not absorbed by the factory? )
[Deleted]  
Aleksey Nikolayev #:
Imho, it will take off if there are modes in the market that are (a) clearly defined and (b) live long enough after they are defined.

Well, they live for a long time, they are just ranges based on some attribute. The TS themselves need to be re-optimised on them periodically.

 
Aleksey Nikolayev #:
Imho, it will take off if the market has modes that (a) are clearly defined and (b) live long enough after they are defined.
It won't take off, it's like a trend - by the time you identify the mode, it's already over.
[Deleted]  
Aleksey Nikolayev #:
Is it already known what island he bought and that it is definitely not absorbed by the factory? )

I think Alyosha was taken into slavery by some big investors, I don't know about Master :)

 

We can assume the existence of time-lagged regimes based on the following assumptions:

A) Big players cannot change the rules of their game too quickly. I mean, of course, they can play as fast as they want, but they do not change some internal rules, regulations on a daily basis.

B) Their constancy affects the price behaviour.

C) There are some predictors that allow to determine quite accurately the time intervals of such constancy.

But this does not justify the mandatory recurrence of regimes, for this we need additional assumptions.

[Deleted]  
Aleksey Nikolayev #:

It is possible to assume the existence of time-continuous regimes based on the assumptions:

A) Big players cannot change the rules of their game too quickly. I mean, of course, they can play as fast as they want, but they do not change some internal rules, regulations on a daily basis.

B) Their constancy affects the price behaviour.

C) There are some predictors that allow to determine the time intervals of such constancy accurately enough.

But this does not justify the obligatory recurrence of regimes, for this we need additional assumptions.

For example, volatility is conditionally constant