I have readed almost half of the thread and I have been trying to create an EA in Matlab based on this concept. I have already implemented different variations mentioned but haven't founf yet a clearly better one, I'll implement a parameter optimization based on Genetic Algorthms with the proper machine learning dataset magement (training to adjust internal paramenters, validation for external parameters and test dataset for independent evaluation). But only after being sure that the system behaves as the reality does.
After three intense days of reading a lot (from other threads too) I finally downloaded Vitrite to see how this works and came up with a scren SEPARATION of 50 pips en EU and around 65 in GU, I like to call it SEPARATION instead of DIVERGENCE because for me they are different concepts. So I opened a small orden in both pairs proportional to the price of the other pair (i.e. I bought 33% more in EU than in GU because GU/EU=1.33 aproximately so from the EU screen (the transparent one) one pips movement in the EU is the same than in the GU). In the optimization phase I will try same lots, proportional as I mentioned now, using Beta mentioned in a lots of post before and co-integration parameters as it should be more neutral in long term. But that's for later.
So as I mentioned, for me DIVERGENCE is other concept, it's one pair making higher high while an indicator (or pair) is making a lower high. I draw some divergences in the 1min betwenn EU and GU and couldn't be more impresed how well they behave. And its made directly from price action! so no LAGGY indicators needed!. So I will altry try to implement a divergence indicator based on two pairs (See first image).
About the normal system I'm having some problems and some feedback would be nice. I'f I use a fixed period windows (like 1200 bars) the price reescales so I can close my positions in lost. If I fix this window after I enter a trade, the price can diverge and take pretty long to close in profit so maybe it's better accepting some lost with a fixed SL but I was also thinking about using some "speed normalizer" so I can use it to avoind this kind of separations or "divergence". In the second image you can see how an opened trade can last for hundreds or thousand of periods, which increases the overall drawdown (see third image). The simulation images are with "random" parameters only for illustrative purposes.
PS: Hope my trade finish in green pips, if not my fault for not trying demo first jeje but i prefer real with small lots.
Divergences and separation trade:
Trade simulation, Blue line is EU, Cyan line is GU. Trades are marken with red and green triangled, red and green line if they are profitable or not.
Return graph. Res line is actual drawdown. Green is acummulated profit of closed trades. Blue line is total ptofir if the trade is closed in that period.