ERUUSD spectra - is this proof of non-stationarity? - page 11

New comment
 
The window problem is most interestingly posed when calculating the Hearst index. The market is thought to have a memory of the news that generated the movement. Different news - different memory length (different window width). Using history, we calculate the window that arose on the previous news and the end of memory. The end of the window gives the Hurst index = 0.5 or <= 0.5, indicating the end of the window and the start of the flat. Exiting a flat is an increase of Hurst (H) more than 0.5. I am curious, at H > 0.5 what is the SPM?
[Deleted]  
Zhunko >> :

There are no sustainable frequencies, of course. But does it really matter?

Of course it is important, otherwise people would not be attracting various optimising techniques for their TS which they are counting on to make a profit at the moment. That is, there is no guarantee of any kind of planned profit, it is also utopia. That is, in fact, all these pams with crazy profit figures are also just a fluke, luck, if you like. But nevertheless it does not exclude the profitability of TC, if one earns little and steadily. In fact it is expressed in sustaining huge drawdowns, in case of triggering of a stop, due to which the deposit as a whole will preserve the possibility of future growth, when the changed market conditions will not force the deposit to call its most faithful friend kolyan for help.

[Deleted]  

faa1947


Thank you. I'll have a look.

 
Urain >> :

Read on here.


Urain >> :

If there are no other ideas for detecting an event (that is the starting point of the event), it is possible to take a zigzag for example.

The spectrum will be searched until a new extremum of the zigzag is set, parameters can be selected with the tester.

Once a new extremum has been set it means a new window and a new search. After all, the spectrum is floating so why should I bother with the one that has already been cancelled?

Before finding the spectrum I recommend subtracting a linear regression with the same window from the extremum to zero from the quotes.

then you'll get around the Kotelnikov-Nyquist theorem.

Thanks for the link. I read LProgrammer's quarrel with Prival or Prival with LProgrammer(both are mostly irrelevant) with great interest.

But only I didn't understand what "Prival-schooled" is and how one can get around the Kotelnikov-Nyquist theorem.

Could you please explain in more detail?

By the way, the Kotelnikov theorem has to do with signal recovery after sampling. We already have a discrete signal.

Why reconstruct it? I think we were talking about measuring the spectrum of that signal. These are different things.


Why linear regression? You want a more stationary result, i.e. to remove the trend.

But that may not be essential for measuring the spectrum. What else will you drop (along with the regression subtraction) from the spectrum you are going to measure?

It will be essential when you are going to use the spectrum you get.

 

Can I just ask?

Should we anticipate Stationarity (or vice versa...) in data series where there is a trend in the mean?

If yes - where?

If not, why?

------ The truth needs to be known - to me.

 
"there's a spectrum of quotes in the window Dude just." (C) SLammer
[Deleted]  
faa1947:

I am attaching the spectra of the quotation ranges for H1. Two sequential in time and then a common one for them. Nothing in common. And that's on a short time frame.

And why is the He axis only limited to 150? And why not try to "impose" spectrograms on each other with window lengths for example from 1000 to 1500 or 1440*2), at the same time to see how the spectrum of bar sizes would look like modulo, so, for clarity, to see spectra from the difference of a constant component and a cotier on the cascade of windows. And for different TFs, multiples for example.
1...4567891011
New comment