ForumSections

Machine learning in trading: theory, models, practice and algo-trading - page 2603

New comment
mytarmailS #:
So what is the obviousness of (1) and what are the arguments for its validity?
Philosophically, by the infinity of time. And also the impossibility to establish the root cause.
mytarmailS #:
So what is the obviousness of point (1) and what are the arguments for its validity?

Knowledge of such ways would allow to create winning strategies (at least by discarding knowingly losing ones) and sooner or later only they would remain in the market (due to reinvestment of winnings, for example). It is impossible to have a market where everyone wins - where will the money come from?

Valeriy Yastremskiy #:
Philosophically, it is the infinity of time. And also the impossibility to establish the root cause.
If the strategy works for 100 years, then in terms of our life this is considered infinity.
Aleksey Nikolayev #:

Knowing such ways would allow creating winning strategies (at least by discarding obviously losing ones) and sooner or later only they would remain in the market (due to reinvestment of winnings, for example). It is impossible to have a market where everyone wins - where will the money come from?

Well, I agree that it's impossible to know with 100% probability that TS will work. But it is possible to predict the probability and shift the probabilities to your side.

If you have 100 strategies and you know that with a 60% probability they will work in the future, you can make money on it.
mytarmailS #:
Well, ok I agree that it's impossible to know with 100% probability that TS will work. But it is possible to predict the probabilities, and shift the probabilities in your direction.

If you have 100 strategies and you know with a 60% probability they will work in the future, you can make money on it.

Then just go to the expectation and we get the same reasoning about impossibility.

Hence, by the way, it follows that the probability theory approach is poorly suited to describe market uncertainty (too crude, or what), if we still want to find a way to make money. I think a game-theoretic approach would be more appropriate -- one can see how one often tries to use itintuitively in reasoning about what and when certain market players do. Unfortunately, game theory is still a poorly developed (for our purposes) part of mathematics.

Aleksey Nikolayev #:

Then we simply pass in the reasoning to the expectation and get the same reasoning about the impossibility.

Hence, by the way, it follows that the probability-theoretic approach is poorly suited to describe market uncertainty (too crude or something), if we still want to find a way to make money. I think a game-theoretic approach would be more appropriate -- one can see how one often tries to use itintuitively in reasoning about what and when certain market players do. Unfortunately, game theory is a poorly developed piece of mathematics for our purposes.

Incorrect formulation of the problem leads to incorrect conclusions...

For example: Determine a pricing algorithm...

In theory, the solution of this problem will give us profit, but the subtlety is that the problem itself has no solution ... because it is tied to the sea of heterogeneous information, which is not possible to track ...


The correct definition of the problem is half the solution!

mytarmailS #:
If the strategy works for 100 years, then in the context of our life, it is considered infinity
No. The stability and duration of the consequence is determined by the properties of the cause. But even having determined the causes, one can estimate their stability and duration only probabilistically. More true than by properties of the consequence, but still not exact. If we are talking about multifactorial processes. And also feedback processes are difficult to take into account.
mytarmailS #:
Well , ok I agree that it's impossible to know with 100% probability that the TS will work. But it is possible to predict the probability, and shift the probabilities to their side.

If you have 100 strategies and you know that with a 60% probability they will work in the future then you can make money on it
So many people work.)
mytarmailS #:
Question : how can you distinguish one rule from another in the trayn, test, crosvalidations... Before the validation phase.
What's your problem with the validation stage?
secret #:
What's wrong with the validation step?
Where did I write such a thing?
1...259625972598259926002601260226032604260526062607260826092610...3749
New comment