Random Flow Theory and FOREX - page 25

New comment
 
Mathemat:

Nah, Prival, Momentum is an indulgence that is calculated by the simplest obviously non-physical formula. Here is the link: https://www.metatrader5.com/ru/terminal/help/indicators/oscillators/momentum. There is also ROC, something similar: https: //www.mql5.com/ru/code/9340 .

About ticks - here's a link to a thread with my attempts at tick research: 'Tics: amplitude and delay distributions', see my last picture on the first page of the thread (tick process for oira). 99.5% of all ticks are +-1 and the rest is unaffected.


Explain the phrase "completely different ? But for momentum, I don't think that's what Kakdid meant (I hope so). Waiting for the author.
 

Well, that's what it says in my comment:

And there is another graph, which is very interesting. These are now the amplitudes of the ticks, but also in order of arrival. Horizontally is the timeline, vertically is the amplitude. Here the situation is almost unambiguous: there is no special time heterogeneity as in the previous graph. 99.5% of ticks are +-1, almost everything else is +-2. The solid blue shading between -1 and +1 indicates precisely the overwhelming incidence of minimal in amplitude ticks. We may assume that the process is almost stationary. <br / translate="no">

This is a week's worth of ticks, i.e. about 24,000. About 200 ticks are almost evenly split - that's +-2. The rest can simply be neglected. The process is almost stationary (in appearance), significant values are +-1, +-2. What does it mean? It means that the integral of this process will be almost Wienerian. Don't forget that we do not take tick lags into account here.

And that lag is the very dog that makes minutiae not a Wiener process at all. Do you see the difference between your Fig. 2 and my figure?

 
Prival:

But I keep wondering what is the signal and what is the noise in this stream. What do you mean by that, because when you build a model there is also the notion of excitation noise (EFN).


I tend to think that this question has no answer by itself. But if we are bound to a certain time horizon of the game, everything that happens at shorter times can be considered noise. Or maybe not everything :). It is known that the averaged spectral density of price charts looks like 1/f. I have made such graphs myself and indeed everywhere, except for the high frequency limit, the conformity to the 1/f law is very good. The deviation towards white noise started at frequencies around 1/3 (1/min), but this is too close to the sampling rate (minute) to be able to say that it is even there.

Yeah about ACF not getting on the screen, try normalization like I did in the 4th picture. You get a constant and when you output "interesting points" you can just take it into account. I do not know, if it is possible to do it in MT4, I can do it in Matkad by easy hand movement :-).

In MQL you could do it too, but it would still look crooked.

Edit. Almost forgot to ask Momentum is an intetion, i.e. is there a concept of mass? If so its calculation options ? Very interesting bouquet may turn out, force is there, energy is there, inertia too, mass is left.

Momentum is a term of tehanalysis and means simply the increment of price for some number of bars. If we ignore division by dt, the increment on two bars will be acceleration, on 4 bars - acceleration of acceleration, etc. Seems to me that this is just where the analogy with mechanics might start to limp.

P.S. No, with accelerations it seems I have fooled :). Momentum is the sum of velocities, not difference. Still, glitches happen more often without paper :)
 
Actually, it's not clear about limping. In principle we can consider momentums as components of the state vector. But then the dimensionality of the problem goes somewhere towards "wow" :)
 

Candid

How interesting is the human brain, seeing the same graph to draw different conclusions ;-).

I thought you saw inertia there. After reading the word momentum, I thought that you associated the correlation time with inertia, in my brain I got an association such as the ball rolls on the floor due to inertia and while its speed is correlated inertia acts (seems almost a direct relationship). But that's why it doesn't roll for 1 second :-), I thought you would tell me.

For the noise, spectrum and tick lags.

Thank you do not give my brain a rest, but that's how it all came together in my head stupid thoughts. That's the lyrics now more specifically

Mathematician

For the lags, IHMO it doesn't matter. I think this picture has misled you with its apparent stationarity, take the sum and you get how the price behaves, only the sampling rate of the process is different you get it in more detail. Check with ACF, it will be the same as on my 4 (Wiener's output on ticks and non Wiener's output on ticks is probably wrong) (yes for tick lag, it took a long time to figure out what this search gives - high quality teak parquet, and only in 1 place of the internet found, your post tick history 1 ms). If I understand it correctly this is the time interval between teak arrivals (correct if wrong).

Now for the sad part.

If we take minutes, then according to Kotelnikov's Theorem minimum period of the process (for example I consider a sinusoid) that we can analyze is 2 min, but in practice sampling rate should be 5 times more (try to see that it is a sinusoid with 2 samples per period, rather a saw). I.e. we get about 10 min, now for reliable detection (recognition) of a sinusoid you need at least 2-3 periods. What do we get as a result of these sad thoughts.

At this sampling rate.

  1. All processes which have a period of oscillation less than 2-5 minutes are noise.
  2. Estimated time of recognition of simple sine wave 20-30 min. ;-(((((((((
  3. The only possible decrease in time of recognition (detection) is transition to tics :-(((((((((( even worse
    4. .

P.S. Here you have the noise, spectrum, 1/f, + lag + stupid delirious brain, can go drink vodka, because I smell a sheet (that would find the error) here I will not get away :-)

 
Prival:

For lags, IHMO it doesn't matter. I think this picture has misled you with its seeming stationarity, take the sum and you get how the price behaves, only the sampling rate of the process is different you get it in more detail. Check with ACF, it will be the same as on my 4 (Wiener's output on ticks and non Wiener's output on ticks is probably wrong) (yes for tick lag, it took a long time to figure out what this search gives - high quality teak parquet, and only in 1 place on the internet found, your post tick history 1 ms). If I understand it correctly it is the time interval between teak arrivals (correct if wrong).

Yes, that's right, from English lag - "delay". About ACF I can say that it's not so simple. I mean the following: no matter how much we try to reduce a real process to a Gaussian (Wiener, Martingale etc.), we will not manage to do it completely.

Well, for example, because, say, the Fibo-ratios between the swings formed by the Zigzag will remain the same (by price, not by "time"), i.e. the bars will still be dependent, although the process will definitely be closer to the Wiener one. You see, identical p.d.f. does not mean the same dependence between the counts.

Concerning my error: when I write an indikator with equivolume bars and construct the p.d.f. of equivolume bars by size, we will talk then. In the meantime, we're just being philosophical here. And the discretization interval (in the sense of Kotelnikoff th.) has nothing to do with it, in my opinion. It is simply a completely different representation of the market process.

 
Prival:

I thought you saw inertia there. I thought after reading the word momentum that you associated the correlation time with inertia, in my brain there was an association like the ball rolls on the floor due to inertia and while its speed is correlated inertia acts (like almost direct connection). But that's why it doesn't roll for 1 second :-), I thought you would tell me.

The question of inertia is not even discussed, of course it's there :). As for the ball, I advise you to read the post that opens this topic once again :)

  1. All processes which period of oscillation is less than 2-5 min - noise.
  2. Estimated time to recognize the simplest sinusoid is 20-30 min ;-(((((((((


So the characteristic times of working models can be from a day or more, the characteristic time of holding positions can be from a few hours? How is that a bad thing?

Mathemat, explain me please, why to reduce the real process to a Gaussian (Wiener, Martingale, etc.) (C Mathemat)

 
lna01:

So, the characteristic times of working models may be a day or more, and the characteristic time of position holding may be several hours? How is that bad?


Bad not holding time positions, and switching speed when changing models (the time required for detection (recognition)) 30 min at a good move is about 60 points, I would like to go faster. And the theoretical limit - the minimum detection time for model changes 2 min, ie when all is perfect, but this as you know does not happen.

Thanks for reminding me of the first page of this thread :-), I have reread it. I would have corrected some things there, not the idea itself, but I would have formulated my thoughts more accurately. It's nice that we have made progress in the investigation, we have managed to do a lot, and the best part is that we have a lot of work to do, all the most interesting things have just begun :-).

 
lna01:
Mathemat, please explain to me why to reduce the real process to a Gaussian (Wiener, Martingale, etc.) (C Mathemat)

Mathematician think will answer himself why he needs it, I think I understand his goal and think that in achieving his noble goal the developers of the terminal (in particular TS-tester) must be interested in the first place :-).

Why should I, I hope WE need to reduce the real process to a Gaussian one. I will try to explain, it seemed simple and understandable to me, I've just stepped on this rake 100 times, all the time I think that you have the same in your head as I do and therefore everything and everyone should be understandable. I apologize for not explaining it earlier.

See what we do with you. 1 Subtract mu from the real flow (straight line equation). We check BGS, no we go further and examine residues (what remains after subtraction, pay attention may be it already = 0). It seems to have found oscillations with definite amplitude and frequency, we have subtracted them from residuals and got residuals #2. Checking for compliance with BGS, let's say yes. Thank goodness hallelujah, we know all components of the process, including noise and signal, and all parameters are known to us. Straight line is clear, oscillation too (their sum is signal) and CGBS (noise is noise), that is not worth studying (investigating). On the contrary, poor Gauss should be pitied, it is rumoured, that as soon as they start to study CGBS he is overturned there :-)

Edit: thoughts are again popping into my head, and maybe go the other way round, start from the tail. We filter the process thinking it is a Wiener (mathematician knows the parameters +-1 pip), but it is not Wiener and we open the trade. I wish I could switch from my NIL to the NIL for learning forex.

 
Prival:

Why should I, I hope WE need to reduce the real process to a Gaussian one.

...

Thank goodness, we know all the components of the process, including the noise and the signal, and we know all the parameters.


So the transformation which reduces BP prices to white noise will be the market model. Now that's what I understand :)
1...181920212223242526272829303132...85
New comment
Reason: