4
$\begingroup$

Cointegration trading question

What's the state of the art when it comes to choosing proper subsets of stocks/assets where cointegrating relationships aren't ignored as (likely to be) spurious?

For example, we wouldn't want to trade a Chinese stock of a fast food franchise with an Australian mining company because a cointegrating relationship could too easily be due to a type I error. But you might want to trade two Australian mining stocks because they share risk factors, reducing the probability that your cointegrating relationship is spurious.

What's the best way to approach this? Informed opinions welcome.

Improve this question
$\endgroup$
3
  • 2
    $\begingroup$ The simplest way to cull possibilities, as you point out, is to look at sector affiliation and geographic region. No need to make things overly complicated until you have a specific goal. $\endgroup$
    – chrisaycock
    Commented Nov 12, 2012 at 20:04
  • $\begingroup$ @chrisaycock Thanks. It's that simple is it? What if you had that all set up, what would be the next step? $\endgroup$
    – Jase
    Commented Nov 13, 2012 at 2:05
  • 1
    $\begingroup$ I second what chrisaycock said, why dont you run some tests and if you are not satisfied with the results, come back, report why and what you have done. I generally only answer questions where the OP has shown he/she has put in some work on his/her own, open ended questions generally do not garner much activity by most who are here trying to help...just my 2 cents $\endgroup$
    – Matt Wolf
    Commented Nov 13, 2012 at 4:44

2 Answers 2

Reset to default
1
$\begingroup$

Simple pairs trading yield is falling every year. It is now at the level when "spurious" cointegration cases become a last resort.

Improve this answer
$\endgroup$
3
  • 2
    $\begingroup$ All the same, the OP seems really intent on trying it. So if you have some advice on things he can try, that would be welcome. $\endgroup$
    – chrisaycock
    Commented Nov 16, 2012 at 1:52
  • 1
    $\begingroup$ I am interested, do you have a reference for this claim? Or at least could you tell me how you came to know this? $\endgroup$
    – Jase
    Commented Nov 16, 2012 at 15:17
  • 2
    $\begingroup$ the reference: 1. gogerty.com/blogpersonal/… 2. more complex and slightly outdated: stat.wharton.upenn.edu/~steele/Courses/434/434Context/… $\endgroup$
    – dkhokhlov
    Commented Nov 18, 2012 at 20:50
1
$\begingroup$

You want to find combinations (2+) of securities with strong economic ties, e.g. WTI/Brent crude, gold/goldMiners, DJIA futures / component stocks, etc.

The theoretical edge would come from the break down in correlation between the combinations in the short run, if their long run correlations remain robust. It's likely that once you understand where the money is, you'd see that cointegration doesn't really add any value.

Improve this answer
$\endgroup$
4
  • $\begingroup$ Thanks Rocko. Can you please elaborate on your last sentence? $\endgroup$
    – Jase
    Commented Dec 16, 2012 at 3:18
  • $\begingroup$ Hi, what I meant is that the number crunching such as cointegration is only curve fitting, which may not present an actual edge that's likely to persist into the future. You need to focus on discovering exploitable economic inefficiencies; the number crunching is really only to see HOW profitable an edge is. $\endgroup$
    – Rock
    Commented Dec 16, 2012 at 8:25
  • $\begingroup$ with humulity i recommend you take care to the way the gold mine is financed before putting your money in a gold/gold mine spread trade $\endgroup$
    – tagoma
    Commented Dec 16, 2012 at 17:06
  • $\begingroup$ I think Edouard just made a good point of something that number crunching wouldn't picked up. Gotta do your homework before going live with any trade. $\endgroup$
    – Rock
    Commented Dec 16, 2012 at 20:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.