Building a trading system using digital low-pass filters - page 13

 
grasn писал (а): Well, take as a basis a series, which is already stationary by definition (if you search - there are many), why do you need these criteria????
Well, I need that this stationary series can be somehow easily obtained from the real market series. Only then, by generating a similar stationary, I will be able to reconstruct something similar to the real rynag...
 
Mathemat:
grasn wrote (a): Well, take as a basis a series that is already stationary by definition (if you search - there are many), why do you need these criteria????
What I need is that this stationary series can somehow easily be obtained from a real market one. Only then, by generating a similar stationary one, I will be able to reconstruct something similar to the real rynag...

Unfortunately, I can not assess the depth of the whole idea, And the point? The only way you can get a stationary series is to remove all the trends. OK, Take the moving MA, subtract it from the price and you get a good approximation of stationarity. You can find the optimal MA window by estimating the resulting distributions. If you don't remember the MA curve, there is no way to reconstruct the original series. You can divide the initial series into segments and remove trends locally; you can do a lot of things.

It seems to me that it is easier to take a ready-made stationary series with the necessary parameters and restore "anything" on its basis. Although, as I wrote above "I can't estimate the depth of the whole idea", maybe I'm wrong and the "native" series must be restored.

PS: I think the main problem is how to generate non stationary. :о(

 

And I liked my idea (you can't praise yourself :o):

Here's another option: why not use sequential generation of zigzags, where each element y = a + b * x, parameters a, b, N (segment length) you set "at random". Plus you impose noise. Necessary distributions can be "looked up" from real zigzags. What's wrong with it?
When I have free time, I'll try it.
 

Sergey, here is the link I posted for Prival: https://forum.mql4.com/ru/9358/page6#51829, my second post on the page. Check it out and you. Questions and suggestions are welcome.

The only way to get a fixed line is to delete all the trends. OK, Take the moving MA, subtract it from the price and you get a good approximation of stationarity.

I'm not so sure about the stationary series there (why MA and not regression?). That's why we need a proper stationarity test.

 
Mathemat:

Well, I need to be able to get this stationary series somehow easily from the real market series. Only then, by generating a similar stationary one, I will be able to reconstruct something similar to the real rynag...

What prevents you from generating a similar non-stationary one?
 

bstone, if it's that easy, generate it and post the result here. And we'll see...

P.S. Only with a more or less rigorous justification, of course.

 
Mathemat:

bstone, if it's that easy, generate it and post the result here. And we'll see...

P.S. Only with a more or less rigorous justification, of course.

Well, I don't have that need right now. It's a waste of time. If you tell me what's bothering you, maybe I'll give you an idea how to make it work. I got time for that.
 
Mathemat:

Sergey, here is the link I posted for Prival: https://forum.mql4.com/ru/9358/page6#51829, my second post on the page. Check it out and you. Questions and suggestions are welcome.

The only way to get a fixed line is to delete all the trends. OK, Take the moving MA, subtract it from the price and you get a good approximation of stationarity.

I'm not so sure about the stationary series there (why MA and not regression?). That's why we need a proper stationarity test.

No, "only robbery and nothing else!" :o). But I wrote about regression too, here:
You can divide the original series into segments and remove trends locally,a lot of things can be done.


Thanks for the link, I read it a long time ago.

OK, I won't distract you with my silly ideas :o)

 
No, you can take your mind off work for a while. Here is some food in the form of a simple, but very funny experiment:

1) We generate a sample of 10,000 values which obey the law of normal distribution in the range [-1;1].
2) Consider this sample as a series
3) Reconstruct the price series by integrating the returns series
4) Plot the graph:


And now let's call the "elliotic experts" from a neighboring branch and be sure they will voraciously argue that we see a classic 5-wave pattern followed by an emerging compound correction X-Y-Z.

Now the question is, if the result is fully consistent with the Elliot wave law (try to pick on it or find fundamental differences from what we observe on the currency charts), how is it not suitable as a synthetic for testing TS?

That's how complicated it seems, but very simple at the same time :)
 

The stage at which I naively thought that returns are distributed according to the normal law has long been passed, thanks to Rosh. And Prival recently, a page or two ago, posted a picture showing that normal distribution doesn't rule here. There is more complicated maths here, with fat tails and sharpest peak in the centre of the distribution.