1 Topic by Irmantas (edited by Irmantas 2016-08-13 12:55:11)

  • Registered: 2016-05-16
  • Posts: 190

Topic: Strategies corelliation test

Hi,
I have system basket, and want to check how much they produce correlated results. Maybe someone has some method to check bunch of exported "Journal" excels for result correlation? Maybe someone can link to some resource or excel file how to do it, if of course someone already walked this path. I would be very great full! smile

At the time most of my found strategies are break out, trend following, meaning that probably all of them will produce wins or losses in the same days, what is very bad. I tried to find them in different pairs, different time frames, but in the news events like yesterday all markets tend to make similar movements... I want to get rid of highly correlated systems, choosing better ones.

Have a good weekend wink

  • footon
  • footon
  • ◄≡≡≡►
  • Offline
  • Registered: 2009-08-17
  • Posts: 2,958

Re: Strategies corelliation test

On top of my head, correlation function in excel. But that applies to 2 data columns at once, so... Hmm, makes me think smile

  • Registered: 2016-05-16
  • Posts: 190

Re: Strategies corelliation test

Today worked hard with excel smile Finished with two columns for excel to calculate R^2 for correlation between 2 systems.
However not very sure should I calculate between these systems balance lines or just for day's end profit/loss? Leaving no traded days empty. What do you think? And what values should be still acceptable? In theory R^2  varies between max 1 (complete correlation) and -1 (negative correlation). I found written that 0.85 is high correlation, but in my case it's totally depends on what I choose to compare. See pic below, maybe it will helps.

  • Registered: 2016-05-16
  • Posts: 190

Re: Strategies corelliation test

Today I finished my portfolio's correlation test, chosen to measure end of the day results with excel function =CORR() smile   However, still can't find answer how much correlation is bad and when I need to remove duplicate system. Probably it must be personal chose and it should depend of systems risk. I guess I am ok till systems produce 60% of similar day end results. After that I think I would delete worse system and would leave better looking.

There is some thing to mention, that correlation is dynamic/changing, and sometimes markets become very correlated, especially in market crashes, so unavoidable similar systems will produce same positive/negative results.

Table of my portfolio system correlations:

And here is example of what end of daily results correlation was measured:

If somebody wants, I can upload this excel file. There is some macros to use to produce data format needed. I just need to clean it a little bit, I am no excel guru smile so please ask.

Have a nice day