Correlation and Cointegration

[holmes16]

hello,

just curious, how many of you correlation traders test for cointegration? i don't trade much forex, but i would have thought that since it's been long established that currency pairs exhibit random walk behaviour, it would be important to test for cointegration before making correlation based trades. thanks,

 

[timsk]

hello,

just curious, how many of you correlation traders test for cointegration? i don't trade much forex, but i would have thought that since it's been long established that currency pairs exhibit random walk behaviour, it would be important to test for cointegration before making correlation based trades. thanks,

Hi holmes16,
I'd say you're correct if you're trading spreads on pairs which have diverted from the mean by X amount and the trade is based on the assumption that the pair will revert back towards the mean. However, there are other reasons for wanting two highly correlated instruments, e.g. in order to use one as a hedge against the other. Then the cointegration aspect of the pair isn't particularly important.
Tim.

 

[holmes16]

hi timsk,

i would argue that testing for cointegration is important for hedging as well. The reason being the following:

The hedge ratio that you would typically use comes from the regression coefficient taken from regressing one pair against another. For the results of this regression to be usable, it is important that neither currency pair have a unit root, or for both pairs to have a unit root and for them to be cointegrated. If this is not the case then the hedge ratio won't be consistent (i.e. the estimated hedge ratio won't converge to the true value).

 

[scose-no-doubt]

Okay my stats knowledge pretty much tails of at regression analysis relatively simple interpretation and have never really tried to apply anything to a real world finance scenario. I'm too busy (lazy).

Anyway, can you please explain to me why you need more than regression analysis to prove correlation? What am I missing 'cos I would have thought that if any unit root issue (a concept which I'm unfamiliar with) had an effect on the series it would have presented itself through heteroskedasticity in the regression analysis itself eh? :s Furthermore, how would you go about testing for/proving unit root in a moving real world current market?

Interesting stuff...

 

[rob the bank]

Okay my stats knowledge pretty much tails of at regression analysis relatively simple interpretation and have never really tried to apply anything to a real world finance scenario. I'm too busy (lazy).

Anyway, can you please explain to me why you need more than regression analysis to prove correlation? What am I missing 'cos I would have thought that if any unit root issue (a concept which I'm unfamiliar with) had an effect on the series it would have presented itself through heteroskedasticity in the regression analysis itself eh? :s Furthermore, how would you go about testing for/proving unit root in a moving real world current market?

Interesting stuff...

What the f**k have you guys been smoking - can I have some please?

 

[scose-no-doubt]

You smowking da weeydz, Robby?!

I ain't smokin' nuttin'!

 

[tradern]

thanks for the pdf.