From theory to practice - page 53

 
Alexander_K2:

Yes, it has already been developed and I ran it today on 4 pairs at once so far.

The question is whether the market process is Markovian or non-Markovian. If it is a non-Markovian one, then historical archive tick data should be taken into account, if it is a Markovian one, then it should not. Well, we're bickering here...


At some interval it may be a "no memory". Then the market changes the character of movement and it turns out that it remembers everything perfectly well, and not only the nearest ticks, but also the states of one year ago.

 
Олег avtomat:

DO NOT MUST.


SO

this t2 allocation is a bit of an obsession...


What a bunch of bullshit!

Incremental modelling is the mainstream of today's econometrics. It's called GARCH. Everything is so well known, it is so well known WHAT is solved and WHAT is not solved - one should just take interest instead of trashing threads.

If about distribution.

Today the incremental model consists of three parts:

  • The trend model - usually an autoregressive model (probably the whole "Markovian-Non-Markovian" thing). This deals with the number of previous increments on which the current one depends. Usually from the previous one. I had a maximum of 5 previous
  • dispersion model. This is some kind of GARCH model that considers various nuances of dispersion.
  • distribution model. Usually different skewed distributions. It is believed (proven on) that the best fit (smallest error) is the Student's skewed distribution. Degrees of freedom are calculated, they are different. But it is also known that incremental distributions change all the time

There is ready mathematics for this, and widespread.


PS.

To take ticks or not to take?

The shallower the TF, ideally ticks, the more accurate the variance values (all statistics in general) and the more accurate the distributions, which never fully coincide with their theoretical ideals. So when talking about ticks it's a matter of accuracy. I don't even need minutes that way. If it's more accurate, you're stacked against the spread or something else.


PSPP

When will they start banning people like this? When will they make it compulsory to review literature? There is this requirement in the articles on this site.

 
Олег avtomat:

At some interval it may be dabbling "in oblivion". Then the market changes the character of the movement and it turns out that it remembers everything perfectly well, not only the nearest ticks, but also the states of a year ago.


Absolutely accurate and long time ago it is called market fractality: it remembers the previous/next tick and also remembers the previous/next M1, M5, etc. and the explanation is very simple: there are investors in the market, working on different horizons.

 
Renat Akhtyamov:
What conclusion have you drawn - Markovian or non-Markovian?
Non-Markovian. Just don't know exactly how to read the ticks, in what sequence. Because if I read them using exponentially distributed time, they suspiciously look like Markov's ones. And to read every tick - I simply don't have enough computer power and VisSim capabilities...
 
Alexander_K2 Source data - EURJPY.zip
You have links in this xls to other xls that are not in the zip. And I am too lazy to process such a long series. First 100 increments - will that work? Just put 100 ticks in a text file.
 
СанСаныч Фоменко:

That's a load of crap, not a branch!

Incremental modelling is mainstream in today's econometrics. It's called GARCH. Everything is so well-chewed, so well-know WHAT is solved and WHAT is not solved - one should just take interest, and not start flooding threads.

If about distribution.

Today the incremental model consists of three parts:

  • The trend model - usually an autoregressive model (probably the whole "Markovian-Non-Markovian" thing). This deals with the number of previous increments on which the current one depends. Usually from the previous one. I had a maximum of 5 previous
  • dispersion model. This is some kind of GARCH model that considers various nuances of dispersion.
  • distribution model. Usually different skewed distributions. It is believed (proven on) that the best fit (smallest error) is the skew Student distribution. Degrees of freedom are calculated, they are different. But it is also known that incremental distributions change all the time

There is ready mathematics for this, and widespread.


PS.

To take ticks or not to take?

The shallower the TF, ideally ticks, the more accurate the variance values (all statistics in general) and the more accurate the distributions, which never fully coincide with their theoretical ideals. So when talking about ticks it's a matter of accuracy. I don't even need minutes that way. If it's more accurate, you're stacked against the spread or something else.


PSPP

When will they start banning people like this? When will they make it compulsory to review literature? There is this requirement in the articles on this site.

Thank you this time, SanSanych! Good comment.

If they ban me, at least I will understand the process.

 
bas:
You have links in this xls to other xls that are not in the archive. And such a long row is too lazy for me to handle. The first 100 increments - will that work? Just put 100 ticks in a text file.
There's about 100,000 I think. Don't be surprised by the values 0.5, 1.5 etc. It's just that - the average value between tick increments.
Files:
EURJPY_2.zip  833 kb
 
Alexander_K2:

Thank you this time, SanSanych! Good comment.

Let them ban me - at least I'll walk away with an understanding of the process.


You're giving up fast, there's still three weeks to go.

 
СанСаныч Фоменко:

Absolutely accurate and long time ago, called market fractality: remembers the previous/next tick and also remembers the previous/next M1, M5, etc. and the explanation is very simple: there are investors in the market working on different horizons.


More generally: a system with multiple levels of hierarchy.

 
Alexander_K2:
Just don't know exactly how to read the ticks, in what sequence. Because if you read it through exponentially distributed time, it becomes suspiciously similar to Markov's one. And to readevery tick- I just don't have enough computer power and capabilities of VisSim...

Why do you need to research potikovo?

in forex the steps are discrete!

there is such a thing as a POINT! right now it's a five digits.

and the terminal changes the price when one point has passed.

so the increment size is almost always ONE POINT. why? because the tick size is ONE POINT.

and your investigation of the increments will show that the most frequent increment length is of this size.

it's only when the market is choppy that ticks can be several points each.



But in a random walk, the increments are always of different sizes.

Reason: