Things are correlated until they are not

General discussions about futures.
Post Reply
Chelonia
Roundtable Knight
Roundtable Knight
Posts: 525
Joined: Mon Apr 30, 2007 3:37 pm

Things are correlated until they are not

Post by Chelonia »

Things are correlated until they are not
Attachments
WTI1.jpg
WTI1.jpg (19.9 KiB) Viewed 5233 times
WTI2.jpg
WTI2.jpg (18.26 KiB) Viewed 5233 times
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1052
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

Post by Chris67 »

great post Chelonia
a harsh lesson for many basis traders to learn I'm sure and a fundamental reason to have a lot of instruments in your portfolio
C
nodoodahs
Roundtable Knight
Roundtable Knight
Posts: 218
Joined: Wed Aug 09, 2006 4:01 pm

Post by nodoodahs »

You've got it exactly backwards. Things are NOT correlated ... until they ARE. :lol:

Seriously though, correlation is not a characteristic of population; correlation is an observation taken on a sample from a population. So rather than worry about whether "correlations change" we should rather first worry about whether what we observed is real, or not. Then maybe spend some time worrying about its persistence ...
Chelonia
Roundtable Knight
Roundtable Knight
Posts: 525
Joined: Mon Apr 30, 2007 3:37 pm

Post by Chelonia »

The correlation was very persistent over the past decades and also very real in my opinion until it was not. However, I think one should not make the mistake to NOT include both contracts into a portfolio because of the strong correlation or to assign a high correlation ratio which could result in having a Crude Oil position and not being allowed to add a Brent Oil position. Instead I opted for treating both contracts uncorrelated but made adjustments in our allocation. So instead of trading (for example) 2 Crude and 2 Brent, I trade 1 Crude and 1 Brent with the same AUM.
So basically I manage the persistent correlation risk, but at the same I do not want to miss out if/when a correlation change occurs.
nodoodahs
Roundtable Knight
Roundtable Knight
Posts: 218
Joined: Wed Aug 09, 2006 4:01 pm

Post by nodoodahs »

Chelonia wrote:The correlation was very persistent over the past decades ...
(1) Turn the data into daily movements were each contract held long
(2) Use the =correl() function to generate a rolling N-day correlation for several N
(3) Graph that over time.
(4) See if you still think it was "very persistent"

I haven't examined everything, but I've yet to examine anything where observed correlations (again, it's a measurement on a sample and not a characteristic of a population) don't change over time, and those changes are significant enough to mess with you if your system uses some compensation method for correlations. On this point I think we agree, observed correlations change over time.

You mentioned several methods of compensation for this, one other systematic method might be to track rolling N-day correlations and assign some cutoff where you include/exclude less liquid contracts (or take some other action based on the previous rolling N-day correlations). This is analogous to using a previous rolling N-day volatility for some of your system parameters ...

Whether correlations are "real" but unstable, or whether we're just fooling ourselves based on sample measurements and a psychological need for analytical masturbation, is an academic question and not a practical one - that's true. But it's a question worth asking.
Chuck B
Roundtable Knight
Roundtable Knight
Posts: 481
Joined: Thu Apr 17, 2003 6:34 am

Post by Chuck B »

As was posted up above, it can make profound sense to trade "highly correlated" assets with the same system (i.e. Brent and WTI). What *really* matters is how "correlated" they are (their performance) when applied to your system (which in most cases means how the individual volatility of the asset "fits" with the exit parameters of your system; granted, said volatility correlation, exit correlation in trends, is a forever changing variable on future data). For example, even though assets have a close to perfect correlation, that is by no means a direct indication of how they will perform with respect to your system's exit.
Post Reply