Not the Grail, just a regular one - Bablokos!!! - page 54

 
Meat:
If there is cointegration at two pairs, then only at a small time interval. Moreover, in most cases, the resulting synthetic is almost the same as the cross between these majors (the difference is insignificant). Is there such a thing as a cross hanging in a stable flat for 3 years?

You have checked. In principle, this small time interval could be quite enough to pull the crumbs off the table. Synthetic and cross are not exactly the same thing because of the weighting factors.

Again, preliminary tests show that there is a grain of truth in it. So far, the task is to formulate conditions to be met by the synthetic. The main task is to find an algorithm for determining the optimal number of bars on which this synthetic is built. In a purely logical way it is necessary to choose pairs and coefficients of their participation in synthetic, so that linear regression of synthetic has zero slope angle and dispersion has minimal value. And, of course, tambourine to keep synthetic coefficients up to the moment of obtaining a signal to close the position.

 

ivandurak:

Meat


You have checked. In principle, this small time interval could be quite enough to pull the crumbs off the table. Synthetic and cross are not exactly the same thing because of the weighting factors.

Again, preliminary tests show that there is a grain of truth in it. So far, the task is to formulate conditions to be met by the synthetic. The main task is to find an algorithm for determining the optimal number of bars on which this synthetic is built. In a purely logical way it is necessary to choose pairs and coefficients of their participation in synthetic, so that linear regression of synthetic has zero slope angle and dispersion has minimal value. And of course a tambourine to keep synthetic coefficients up to the moment of receiving a signal to close a position.

Meat

I couldn't agree more. I think in my Cointegration thread I gave unit root tests for 6700 H1 candles. Only occasionally was the residual not stationary. I can try it again if you like.

ivandurak:

the selection of the number of lags is done automatically, and there is quite a wide choice among algorithms to determine the number of lags.

 
ivandurak:

The number of lags is selected automatically, and there is a wide range of algorithms for determining the number of lags.

If it's not too much trouble, please send a magic paddle in this direction.
 
ivandurak:
If you don't mind, plz magic kick in this direction.

Initial quotes, H1 6736 bars, year.

Residuals graph from cointegration regression

Testing for non-stationarity. The numbers on the left are, roughly, the probability that the residual is non-stationary.

One of the outliers = 2.4% is the probability that the residual is non-stationary.

Exactly, we cannot reject the null hypothesis that the series is nearly 100% stationary!

 
ivandurak:

You have checked. In principle, this small time interval could be quite enough to pull the crumbs off the table. Synthetic and cross are not exactly the same thing because of the weights.

Only what does "crumbs from the table" have to do with it if the graph shows a steady increase in equity over 3 years? That is, as if there was a stable cointegration. But it cannot be true if we deal only with two forex pairs. You should not believe in fairy tales. Or have you forgotten that forex is the most efficient market? Even on a bunch of pairs something is unlikely to work out.

There was a man here before under the nickname hrenfx (now banned). He had been studying all that very deeply for a couple of years, tried to create a stationary process out of a lot of instruments (and not only Forex), thus he found some correlations between them and created the Recycle indicator. I managed to find such coefficients that it all looks very impressive, as if the entire market is arranged in folders. But in reality it is not so bright, because coefficients are constantly changing. His subsequently opened pamm account, although generally profitable, has been extremely unstable.

 
faa1947:

...

Precisely, we cannot reject the null hypothesis that the series is stationary at nearly 100%!

And what ratios did you have for EURUSD and GBPUSD? Try to similarly check the roots for EURGBP. And also try moving the interval in question a few bars forward or backward - will there be any differences?
 
Meat:
And what ratios did you have for EURUSD and GBPUSD? Try similarly checking the roots for EURGBP. And also try to move the interval in question a few bars forward or backward - will there be any differences?

The plots were obtained by shifting the window of 236 bars (two weeks) along a sample of 6,736 bars. The unit root test values were recorded in the last bar of the shift window. This is how many test values are obtained.

The quotients themselves are not stationary. That's the point of cointegration, that the combination of two non-stationary quotes is stationary.

 

Meat:
А какие коэффициенты у вас были к EURUSD и GBPUSD?


From regression. There are subtleties there. Doing it the way a bunch of indicators for pair trading are done will not work. The residual will not be stationary and nobody checks it.
 
faa1947:
From regression. There are subtleties there. Doing it the way a bunch of indicators for pair trading are done will not work. The residual will not be stationary and nobody checks it.
And what ratios did you obtain in this interval? We are interested in the specific values, because we are not trading regression, but currency pairs.
 
Meat:
And what are the coefficients at this interval? We are interested in specific values because we are not trading regression but currency pairs.

Hm.

What is a regression?

For example, EURUSD = a+ b*GBPUSD

are evaluated by a and b - that's the coefficients for you. So regression has a lot to do with it.

Look here. There is much more on this subject.