Quantifying ways that stuff moves together and stuff

Henrik · Post by **Henrik** » Tue Jan 10, 2006 12:47 am

Interesting thread indeed.

Another thing, although further away from the subject, you may be interested in bringing up on (or ahead of) the presentation is the way you measure the series, ie the length and density of the data collected.

The correlation between two time series can be very different depending on your approach. This might be obvious to you guys but personally I think it is very important.

This can also easily be abused by less serious sales guys preparing a sales pitch.

JabairuStork · Post by **JabairuStork** » Tue Jan 10, 2006 1:55 am

Maybe I'm a little late to this party, but having been on both ends of alternative co-movement metrics, here is what I think:

Do not use higher momements of the joint distribution if there is any way to avoid it. Even the most robust and large financial data set is likely to lose precision too fast to make this useful, and it has the added effect of alienating just about every potential user.

Do use a method that is directly informed by the ultimate application.

For example, if you are trying to get some super extreme beyond-VaR risk estimates, look into extreme value theory for joint distributions, and potentially non-gaussian copulae (although these are also of questionable accuracy is most financial conexts).

If you are trying to plug this into a version of a quadratic optimization routine, use covar (correlation), possible with some markov-chain induced regime shifiting in the correlation matrix. Variants of this approach are used by some well known consultants and off the shalf optimiztion and risk software providers.

Good luck.

NIP247 · Post by **NIP247** » Thu Jan 12, 2006 5:45 pm

Now that I'm leaving the cosy confines of dreaming up trading projects for a totally different area, I'm just wondering if some of our resident dax-30 mean reversion trading gurus could say if the project was totally muppetesque (in which case I will change my icon to Kermit):

With a trading horizon of around 12 hours, would you right now (although the market just closed), all else equal, have the following positions:

Long TUI and DTE and maybe BASF, short -MUE, -SAP, DP and SCH (in sizes of 1x, 2x, 1x, 0.3x, 0.7x, 0.6x, 0.6x) ?

Maybe this doesn't make sense, in which case I will clarify if there is interest.

FDAXHunter · Post by **FDAXHunter** » Thu Jan 12, 2006 5:57 pm

Why are you long BASF? I don't think that one should be in there. Also, SAP should have been kicked out the selection list entirely because of yesterday? Teething pains?

NIP247 · Post by **NIP247** » Thu Jan 12, 2006 6:17 pm

I generated the list in purely systematic fashion. The signal in BASF (i.e. excursion) is too small, so would probably not include it. Would you filter out SAP because of the warnings/downgrades/removal from focus list etc.? Would you otherwise say that the methodology achieving this list is of some interest?