Spread Correlations

maninjapan · Dec 18, 2013

Does anyone know if there is a 'correct' or 'standard' way to calculate correlations for spreads? (referring to calendar and inter-product futures spreads).

I understand that correlation of returns rather than the actual prices is the standard method for calculating correlations for outrights, however this doesn't work work for spreads....
Perhaps I need to go back to the individual legs of the spread and calculate the difference in changes there?

Any advice or suggestions would be much appreciated.

dunleggin · Dec 20, 2013

You're talking about calculating the correlation between two spreads, yes? (Returns rather than raw price, yes) What exactly do you mean by stating the correlation "doesn't work" ?

xandman · Dec 23, 2013

I think your going too deep into the rabbit hole.

The beauty of Spreads is that they are correlation trades already which immunizes you from one ( or a benchmark) asset class.

Why would you trade the correlation of correlations? Trying to hash out a grand diversification concept? Look up cointegration on Wikipedia. Better automate if you want to trade like that.

maninjapan · Jan 6, 2014

Quote from dunleggin:

You're talking about calculating the correlation between two spreads, yes? (Returns rather than raw price, yes) What exactly do you mean by stating the correlation "doesn't work" ?

Sorry if I hadn't explained myself clearly enough.

I am talking about the correlation between at least 1 spread and a second product, which may or may not be a spread.

A couple of examples:
- 2nd/3rd month spread on CL futures against the front month contract,
- 2nd/3rd month spread on CL futures against the 2nd/3rd month spread of brent futures.

I usually use the close to close change as a percentage to calculate the correlations for stocks or indices, futures contracts etc, however with Spreads, I find the change as a percentage of the closing price to be meaningless when the closing price is the difference between 2 products...

maninjapan · Jan 6, 2014

Quote from xandman:

I think your going too deep into the rabbit hole.

The beauty of Spreads is that they are correlation trades already which immunizes you from one ( or a benchmark) asset class.

Why would you trade the correlation of correlations? Trying to hash out a grand diversification concept? Look up cointegration on Wikipedia. Better automate if you want to trade like that.

xandman, I'm not exactly looking to trade the correlation of correlations as such,but I do like to keep an eye on correlations between different products and am looking to add spreads to the products I track.
I am aware of cointegration, and have looked into it previously but that opens a whole new cans of worms Im not quite ready to deal with just yet.

dunleggin · Jan 6, 2014

My first thought was whether you are calculating the percentage changes of negative spread values correctly (though that wouldn't be the issue in your crude oil example if you use e.g CL2 - CL3 to derive the spread price) .... e.g. in Excel, instead of the normal formula =A4/A3-1 or =(A4-A3)/A3, you'd need to use =(A4-A3)/ABS(A3)

Can you post a snippet of an example of the 'meaningless' results you get for one of your examples? Otherwise, I don't see immediately why your specified approach isn't 'working' .

I agree that correlations of spreads have their place - your example of the WTI/Brent box is a good one, and then there are others such as mapping, say, red/green Eurodollars to the 2/10 Tsy spread for a curve proxy.

I've read some opinions that if a series is cointegrated then one can use correlation of prices rather than returns (and obviously that would apply to the CL spreads) .

Soon2Bgreat · Jan 6, 2014

Quote from maninjapan:

Sorry if I hadn't explained myself clearly enough.

I am talking about the correlation between at least 1 spread and a second product, which may or may not be a spread.

A couple of examples:
- 2nd/3rd month spread on CL futures against the front month contract,
- 2nd/3rd month spread on CL futures against the 2nd/3rd month spread of brent futures.

I usually use the close to close change as a percentage to calculate the correlations for stocks or indices, futures contracts etc, however with Spreads, I find the change as a percentage of the closing price to be meaningless when the closing price is the difference between 2 products...

Try using changes in ticks to calculate the change in $ values - that's a way of normalizing across products and it can be easier to identify how you want to weight the spreads.

ltn2012 · Jan 7, 2014

Everybody spreads, every order that come into the pit is a spread....everybody trade price relationships!

Charlie Difrancesca

maninjapan · Jan 7, 2014

Quote from dunleggin:

My first thought was whether you are calculating the percentage changes of negative spread values correctly (though that wouldn't be the issue in your crude oil example if you use e.g CL2 - CL3 to derive the spread price) .... e.g. in Excel, instead of the normal formula =A4/A3-1 or =(A4-A3)/A3, you'd need to use =(A4-A3)/ABS(A3)

Can you post a snippet of an example of the 'meaningless' results you get for one of your examples? Otherwise, I don't see immediately why your specified approach isn't 'working' .

I agree that correlations of spreads have their place - your example of the WTI/Brent box is a good one, and then there are others such as mapping, say, red/green Eurodollars to the 2/10 Tsy spread for a curve proxy.

I've read some opinions that if a series is cointegrated then one can use correlation of prices rather than returns (and obviously that would apply to the CL spreads) .

If the recent 3 days close for CL was 105 , 106 , 107 then the changes would be 0.95% and 0.94% over the last 2 days
however, if the 1st-2nd month spread closes for the last 3 days were -0.1 , 0 , 0.1 then the % changes for the last 2 days would calculate as 100% and #DIV/0! ( cant divide by 0....) , therefore my comment that these dont 'work'.

Does that make sense? (If not Ill try again)

I have a couple of alternate ideas to try now though so I'll see how I go with these.

maninjapan · Jan 7, 2014

Quote from Soon2Bgreat:

Try using changes in ticks to calculate the change in $ values - that's a way of normalizing across products and it can be easier to identify how you want to weight the spreads.

Thanks for the idea, I will see how this compares to what I have so far.