From theory to practice - page 566

 
Novaja:

So explain how to count?

:))) In Excel, for incremental modules, SUMM(A1:A1440) is 1 value, SUMM(A2:A1441) is 2, etc. There can't not be a normal distribution. You are looking for it, aren't you?

Prepare your pockets, dear Novaja, while I dust my purse...

 
Yuriy Asaulenko:

So, the problem is in the PF, but in detecting a change in the low-frequency part of the spectrum). Moreover, the low-frequency part does not appear immediately, but as the pulse length increases. It cannot be detected immediately by any tricks, neither PF, nor ACF, nor anything else. We can only analyze the difference in levels per dt and consider it a trend when the threshold is crossed. But this will only work as an additional method.

To detect low-frequency components you use low-pass filters - LPF. This is the most reliable way. The best of the simplest is the EMA, which has the lowest phase and group delay. The T parameter in the EMA is a guide and makes no real physical sense. If you do not have a correct LPF, you can only find the EMA, which movement corresponds to YOUR idea of the trend, experimentally on the graph. Next, we differentiate, which cuts off the zero harmonic and oscillating low-frequency components, and get a trend indicator. The level of the octa zero corresponds to the current trend rate. That's it.

1) Let us not offend EMA: in the mql version it is reduced to one parameter (T - which has no physical meaning), but in the applied economic application EMA is slightly more complex and involves in its calculations exactly the desired (wanted, though, want) scenario(marketable) development, which has been coaxed here through cats and other gravitsaps. And the criteria for selecting the initial data to be calculated and the evaluation of the results are given. Since, at the dawn of economic analysis and industrialization, there was a lack of sufficient computing power, simplified versions of xMA, etc. were available, which could be estimated on the accounts or with a pencil in hand.

2) As long as there is no basic version of EMA in a traditional economic primer, discussion of LPF/HF/etc. in this thread will get you nowhere. At least keep the chart scale at 50 pips (for eurusd m1) or 100 pips on the mql grid to see the size of worries immediately

3) And let's start writing the text from right to left in Arabic script? Why? Or like hieroglyphs from top to bottom, a cat hoku by Schrodinger ? If you know mql and know how to use MATLAB or something else, cool, try to make a terminal chart, it's easy, just use Pint to display it horizontally and everything is fine. If you think it's a wild requirement, ok, it's a mql-related resource, use mql and do your studies there or go to a professional matlab meeting, I think nothing has changed in this world and there are enough matlab-based economists:

display horizontally

If we look at the drawing the question: since 9 o'clock the movement has already started and the boundary lines have tapered, in such a taper (as on the picture) the movement can easily go ***a micro rollback and trying to survive in this scale will lead to a loss.

The primitive M1 ema 5 open, channel 0.05 233 of ema 5, bollinger 233 of ema 5, large ema 1440, strategies can be made up without any pseudo-scientific bullshit, if you look at the micro rollbacks, someone is actively living on such strategies:

option

 
Alexander_K:

Ahem... Take a sliding window of 1440 values CLOSE M5 and at each new bar count the sum of incremental moduli. There must, just has to be a Gaussian distribution for such sliding sums. And in the case of periodic ACF (and not only), as Kolmogorov bequeathed, this process is revealed by a neuronet.


EURUSD chart, M5, 2018.09.12 21:15 UTC, InstaForex Group, MetaTrader 4, Real

Where should ACF be attached?

SZY: I'm going to bed, it'll take me 15 minutes to figure out what the customer wants, but what he wants can be figured out for days... nothing changes.

Files:
 
Igor Makanu:

what he wants can be found out for days... nothing changes.

Histogram! If there is a normal distribution - grail, no - yes, nothing changes...

 
Igor Makanu:


And only when it is clear that there is a normal distribution, these amounts should be fed into the neural network.

This indicator is not needed at all - all the work is ahead, it is just preparing the data for the grid and nothing more...

 
Alexander_K:

And only when it is clear that there is a normal distribution, these amounts should be fed into the neural network.

This indicator is completely unnecessary - all the work lies ahead, it's just preparing the data for the grid and nothing more...

Alexander, have you even heard anything about neural networks apart from the name?

 
Yuriy Asaulenko:

Alexander, have you ever heard of neural networks other than their name?

What is it? :))) What's this rural habit of asking questions? I don't want idiotic questions, I want answers!!!

The NeualNet VisSim module does Kolmogorov regression, that's all you need to know.

 
Alexander_K:

What's that? :))) What's the rustic habit of asking questions?

The NeualNet VisSim module performs Kolmogorov regression, and you don't need to know anything else.

It's great, they are everywhere. (I won't interfere further on, let's see what will come of it.) Although I already know - nothing.

 
Yuriy Asaulenko:

That's great, they're everywhere. I won't interfere further, I'll see what comes of it.) Although we already know - nothing.

Hmm... I have to say, I wasn't expecting to throw myself at neural networks...

On the contrary, it's from people like you, Yuri, that I expect to hear: "There is no Gaussian distribution, there is and will never be stationarity and Kolmogorov's theory (remember, I attached the book?) will never work". That would save a lot of time...

However, I heard this phrase already from Automat... Yes, yes, I remember...

 
Alexander_K:

Hmmm... I have to say, I wasn't going to throw myself straight into neural networks...

On the contrary, it's from people like you, Yuri, that I expect to hear: "There is no Gaussian distribution, there is and will never be stationarity and Kolmogorov's theory (remember, I attached the book?) will never work". That would save a lot of time...

However, I heard this phrase already from Automat... Yes, yes, I remember...

The stationary is there. You're looking in the wrong place.) And whether there is a distribution of G or not - it doesn't bother me.