A formal definition of a master-slave - is there one? - page 2

New comment
 
Reshetov:

Correlation coefficients for 1 bar are not calculated, but are calculated for N bars. That is why the response of the slave pair to the movement of the master pair is obtained for N bars history section, not for 1 bar.

There will be noise in any case, but if the correlation coefficients and the less difference for correlation coefficients for pp. 1 and 2, the higher the noise value.

The response time still matters. If the offset is taken - 1 bar, and the pair reacts in 0.9 bar time, that's one thing, it's realistic to get out from under the noise. But if the reaction time is 0.01 bar, that's another thing. Besides this parameter may well not be constant, and the problem is that we don't know it, just like we don't know the noise level... But, as always, practice is the criterion. We have to check it out.
 
alsu:
The reaction time still matters. If the offset is taken - 1 bar, and the pair reacts in time of 0.9 bar, it's one thing, it's real to get out from under the noise. But if the reaction time is 0.01 bar, that's another thing.
Violet. It is still a question of identifying the master-slave. It is not realistic to identify one with a response time of less than 1 bar because the ticks on the different pairs are not synchronous. Only bars are synchronous and only when there are no missed bars.
alsu:


But, as always, practice is the criterion. You have to test it.

Undoubtedly.
 

I want to show you a strategy by example (two signals today, manual play):


 
What if this is the last post here?
 
Reshetov:

There is a method. Everything is learned by comparison. It involves calculating correlation coefficients for two cases:

  1. The first pair is taken unbiased, and the second pair is taken shifted relative to the first pair by 1 bar deep into history
  2. The first pair is taken shifted relative to the second one by 1 bar deep into the history and the second pair is taken unbiased

Compare the correlation coefficients for the absolute values obtained in step 1 and step 2:

  • If the absolute value of the correlation coefficient for p. 1 is greater than for p. 2, then the second pair was the master and the first was the slave.
  • If the absolute value of the correlation coefficient for item. 2 is greater than for item. 1, then the first pair was the master and the second was the slave.


It is not the correlation that should be measured, but the cointegration, and this works if both pairs have the same level of integration.
 
wmlab: I want to show you a strategy by example (two signals today, manual play):
You are taking the meaning of master or slave too literally. It's much more subtle here....))) And the direction is right ))))
 

I agree with the 1st part of the picture, euro went up half an hour early (my time = MSc -1)

Second hike with the pound leading - I don't see it, the euro rolled back almost to the baseline and went back up again

We don't have what to measure the "strength" of the presenter in

As the simplest - number of "steps" - size + speed of "steps" x time

1 "step" is 10% of yesterday's range (euro = 11p, pound = 6p)

 
faa1947:
One should not measure correlation but cointegration, and this works if both pairs have the same level of integration.
We have to check whether the currencies are in long-run equilibrium otherwise speaking, two non-stationary processes, give a stationary linear combination.
 
orb:
We need to check whether currency is in long term equilibrium otherwise speaking, two non-stationary processes, give a stationary linear combination.

Naturally.

If we talk about outperformance, it has a place and Reshetov wrote how to check, so they do. It is used in the stock and commodity markets. There are macroeconomic time series leading. In forex I think Asians are ahead of Europe and states. Australia is good as a lead.

 

There's been talk about correlation... what method do you use to measure it?

There are many, not all of them are suitable.

12345678910
New comment