Market phenomena - page 11

 

Farnsworth:

My understanding is that it is a myth (to say the least), yes, quoting is a complex but not random process (this is really encouraging), there are very strong non-linear relationships. But this does not change the main thing - the price trajectory (what the lines, levels, grids, arcs, etc. are applied to) actually does not characterize the process itself in any way. In other words - you can't graphically find these dependencies.

I agree! I support you!

I'm going a slightly different way, I've already said it before:


And there are already some shifts ;) For example, the structure of the process is already quite clear:

 
ZetM:

It's summer, I live by the sea, it's beach season, the "philosophical holiday" continues....)))


Especially happy for you. And in general, "wave theory" should be studied on practical material - at the waves :o))). I hope you're not building zig-zags in your mind when watching the tide?

I see...)) You are a clever man. Therefore, it is impossible to drive you into a corner, and there is no need to try to do it, it is safer for yourself...)))

I'm fluffy and white and kind, yes, - white and fluffy and kind - all of this is me :o)

 
TheXpert:
Extremes, say.
Maybe we should ask Alexey to tell us more about the "phenomenon of long-term memory" and its (phenomenon) detection methodology? - Alexey, please!, so that it would be possible to move on to the next stage - to think, how to use this phenomenon with advantage. ;)
 
avtomat:

and there have already been some developments ;) For example, the structure of the process is already quite clear:

not everything is clear, at least to me. Still, we need some clarification, at least a little, of who is who on the graph

y-regular process

Is it really possible to isolate the regular process? Or is it just "assigned"?

 
joo:
Maybe we should ask Alexey to tell us more about the "phenomenon of long-term memory" and the methodology of its (phenomenon) detection? - Alexey, please!, so it would be possible to move on to the next stage - to think, how to use this phenomenon with a purpose. ;)


The phenomenon of long-term memory has another "angle" - it is the long tails of the distribution itself. In very crude terms (not even "near-scientific") - all memory sits in these very long tails. I.e. the occurrence of (some given) evasion event (there are subtleties in calculation) is significantly increased for a process with long tails of increments (as opposed to).

PS: trajectory deviation is the sum of increments on a segment of a given length

 
Farnsworth:

Not everything is clear, at least to me. But it does need some clarification, at least a little, as to who is who on the graph

Has the regular process really been isolated? Or is it just "assigned"?

In addition to the previous figure - such a separation with control.

The task is only half done, but already at this stage the result is encouraging ;)

 
avtomat:

In addition to the previous picture, this is a controlled extraction.

The task is only half done, but already at this stage the result is encouraging ;)


there are three lines, red, blue and grey - remind me who they are.
 
Farnsworth:


To put it very crudely (not even "near-scientific") - all memory sits in these very long tails. I.e. the occurrence of a (given) deviation event is significantly increased (there are subtleties in calculation) for a process with long tails of increments (as opposed to).

PS: traction bias is the sum of the increments on a stretch, of a given length


Sergey, do you have specific studies with confirmations (preferably on mql4)?
 
Farnsworth:

there are three lines, red, blue, grey - remind me who they are.

x - grey

y is red

u is blue.

 

My 2 kopecks.

The topic of long-term memory is interesting, in particular for neural network enthusiasts. As for the selection of relevant bars, which somehow influence the zero bar, this is an interesting topic, but it is also exceptionally complicated. In idea, this memory may be trivially insufficient to build a model with satisfactory accuracy.

But if we experiment with synthetic time series, in which this very long-term memory could be measured and controlled, then we could make a neural network forecast on synthetics and observe the obtained accuracy. And then I will do my best to work with quotes, if the memory found in the synthetic is comparable to the quotes, in which case neural forecast is worthy of attention and further time spent.

I'll think about it.