Machine learning in trading: theory, models, practice and algo-trading - page 1578

 
Boris:

That's right, no one promised to converge to zero

The question was about how to trade a mixture of divergent time series, and which one would be at the bottom or at the top we don't know

If you don't know, there is only one way - at random.)

 
Boris:

That's right, no one promised to converge to zero

The question was about how to trade a mixture of divergent time series, and which one would be at the bottom or at the top we don't know

As for the picture: those that were at the top with a greater likelihood remain at the top.

abstractly - consider the Levenshtein distance on the order of succession. The measure is more reasonable - it means the rows are "intertwined" and it is necessary to recalculate them.
If the rows are not taken from spit, then it means that not only you, so many will change estimates and make entries/exits. And here already something can be invented :-)

 
Maxim Kuznetsov:

The picture: the ones on top are more likely to stay on top.

In abstract, consider the Levenshtein distance on the order of succession. The measure is more reasonable - it means that the rows are "intertwined" and it is necessary to recalculate them.
If the rows are not taken from scratch, it means that not only you, it means that many will change estimates and make entries/exits. And here you can already think of something.

They don't even know such words in our village.

Could you make your point clearer?

There is a probability of course and even some statistics... But it cannot help us for some reason ((

 
Boris:


Another hard question for the venerable dons

for example, there are 10+ time series, coming out approximately from one point

how is it possible to trade on divergence of the sum of series, if we do not know beforehand, which BP will be above or below the others?

You have a lot of similar rows, maybe they should be checked for cointegration
 
Maxim Dmitrievsky:
You have many similar rows, maybe they should be checked for cointegration

series can be very similar, this is true

and, I admit, some may even be "cointegrated," although their difference is not very similar to a stationary process

some may go side by side, intertwine, diverge, and converge

the only reliable fact is that they mostly diverge, i.e. the initial distance between the series is smaller than the final distance

of course, if we knew which of them would be higher and lower at the end of the period, there would be no question

 
Boris:

series can be very similar, this is true

and, I admit, some may even be "cointegrated," although their difference is not very similar to a stationary process

some may go side by side, intertwine, diverge, converge

the only reliable fact is that they are mostly divergent, i.e. the initial distance between the series is less than the final distance

of course, if we knew which of them would be higher and lower at the end of the period, there would be no question

Well if the distance between them varies linearly, that's not a problem. It's looking at the picture, I don't know the other one. You can allow buying for one and selling for the other, signals on the balances.

i.e. this is the only option.

 
Maxim Dmitrievsky:

Well, if the distance between them changes linearly, it's not a problem. It's looking at the picture, I don't know the other one. You can allow buying for one and selling for the other, signals on the balances.

So this is the only option.

linearly? what do you mean?

such pictures spoil my mood


and that would be ok, but there are complex overlaps (((



And then there is this.





normal (standard profit) position - when the 1st row - the blue line is located below the others

But it's not always like this, and I can't figure out the reason.

 
Boris:

linearly? what do you mean?

I'm spoiling the mood with pictures like these

and that would be all right, but there are complicated overlaps (((

And then there is this

the normal (standard profit) situation - when the 1st row - the blue line is located below the others

But it doesn't always happen that way and I can't figure out the reason yet

when Y changes proportionally to X, i.e. there is a linear dependence + cycles + noise, it makes no difference which way both crooked lines go, even in different directions

Well, there are not enough observations, it would be 1000 times more. Or to assess the fundamental law behind the curves, how much it is consistent.

That's it, I'm off the forum. Happy New Year!

 
Boris:

linearly? what do you mean?

The mood is spoiled by pictures like these


and that would be okay, but it's complicated (((



And then there is this





the normal (standard profitable) position is when the 1st row - the blue line is below the others

But this is not always the case and I can not yet identify the cause

Boris, synthetic is only the sum of financial series. The only thing you can do on synthetic is indefinitely to narrow the variance. Hence I suggest to continue observing the diverging curvatures not synthetic but original series. You will see the same picture, with the same expanding curvatures.Alas, I can't tell you how to trade this expanding pattern. But I can suggest that if some part of the curvatures began to change the directional vector, most likely some other part should also change its vector of directionality.
It would be more interesting to find out when this expanding cloud of curves has a strong offset from center, then it would be possible to collect statistics and try to identify something useful. But it's not IOC, and therefore I apologize for throwing in off-topic information.