Machine learning in trading: theory, models, practice and algo-trading - page 3102

New comment
 
sibirqk #:

Imho of course, but if you build pricing models, in which there are clear deviations from the SB, then you can then, for example, on this basis, generate artificial quotes, even for a thousand years. Then on this quotation learn with the help of MO to determine those places where there were deviations, and then try to do the same on real quotations. Alternatively.

Deviations from the cb will do what? If it is another stochastic process, it is also random. Randomness doesn't end at SB, it just begins.

In my opinion, these topics have been raised here 100500 times already, and nobody has done anything in this direction.
 
Aleksey Vyazmikin #:

Average, subtract and divide :)

In general, as I understand, you propose to change the target on the section where the signal is "bad"?

At least yes, if we try to equalise through the model.
 
mytarmailS #:
That's what Alexei Nikolaev said.
What do you call this approach?
Well, probably the search for market inefficiency.
 
Maxim Dmitrievsky #:
Deviations from cb will do what? If it is another stochastic process, it is also random. Randomness doesn't end at the SB, it just begins.

I think these topics have been raised here 100500 times already, and nobody has done anything in this direction.
Deviation from SB - for example SB with demolition, and this is already a trend in terms of trading. But I guess you are right, the topic is off-topic for this thread.
 
sibirqk #:
Well, probably a search for market inefficiencies.

I meant whether there is such an approach in official science, as I have already heard exactly the same thoughts about the comparison with the SB

I wondered if there were any established techniques.


Here's a sketch.

on the left is a real chart of the euro m5

on the right SB ticks (cumulative sum) converted to m5

 
Maxim Dmitrievsky #:
At least yes, if you try to equalise through the model.

This may improve results in training, but not in application, if the first plot (with changed target) will appear more often on the history than the second one. But, I need to make the two plots equal so that the model separates them and switches between them, i.e. in theory there should be a categorical feature.

 
Aleksey Vyazmikin #:

This may improve results in training, but not in application, if the first plot (with modified target) appears more often in the history than the second one. But, I need to make the two plots equal, so that the model would separate them and switch between them, i.e. in theory there should be a categorical feature.

Well, you can make up a lot of things as you go along. Then something else will be added, and so on ad infinitum.
 
Maxim Dmitrievsky #:
Well, you can make up a lot of things as you go along. Then something else will be added, and so on ad infinitum.

Of course - there's no limit to perfection!

 
Aleksey Vyazmikin #:

Of course - there is no limit to perfection!

To the madness of the brave :)
 
mytarmailS #:

I meant whether there is such an approach in official science, as I have already heard exactly the same thoughts about the comparison with SB

I wonder if there are any established techniques.


Here's a sketch.

on the left is a real chart of the euro m5

on the right SB ticks (cumulative sum) converted to m5

Visually the graphs are similar.))

Heteroscedasticity is modelled in econometrics and all kinds of applied statistics. There are a lot of tests there. R should have them all. The problem is that they give an estimate of the past and it is not certain that it is suitable for the current moment.

1...309530963097309830993100310131023103310431053106310731083109...3526
New comment
Reason: