negative correlation

Discussions about Money Management and Risk Control.
Post Reply
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1046
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

negative correlation

Post by Chris67 » Tue Dec 23, 2003 12:37 pm

Dear All,

If 2 instruments have a lon term negative correlation .. say -0.85 , are they considered correlated if one gets a signal to buy both ? this happens to me a lot .. where as some may say this is ok as it probably reduces risk ,,, i feel it limits profitability .. maybe simply take the first signal only !!!

regards

d-b
Full Member
Full Member
Posts: 20
Joined: Thu Apr 24, 2003 9:26 am

Post by d-b » Tue Dec 23, 2003 3:53 pm

Chris,

By long term negative correlation, I take it you are calculating correlation in terms of price movement.

I think it depends on how we measure correlation. As far as it is valuable to us as traders, I think correlation should be considered in terms of the likelihood of the outcomes of trades in different instruments being different.

While prices in two markets might be shown as somewhat correlated (either positively or negatively), the more valuable question to ask might be “are the two markets experiencing different market conditions at any given time?â€

William
Roundtable Knight
Roundtable Knight
Posts: 238
Joined: Sun May 04, 2003 4:41 pm
Location: Manhattan, New York

Uncorrelated vs Not Correlated

Post by William » Tue Dec 23, 2003 4:39 pm

David,

Funny, this post couldnt be better timed, i was messing with some numbers today. I want to clarify the terms un-correlation vs. not correlated because it may be causing some confusion for me. (For me, non-correlation means negative correlation, not correlated means 0% correlation.)


100% Un-correlation is = +1, +2, +3 vs -1, -2, -3. Those two number strings are 100% un-correlated. For trading purposes, the US dollar index and Australian dollar are around -90% uncorrelated. (Data from CSI Data's correlation study) So when the Dollar Index trades down, the Australian dollar generally trades up in equal magnitude. To me these instrument's prices movements are very correlated. Dollar up trend = Aussie down trend but choppy dollar period = choppy aussie period, which negatively impacts the smooth return benefit principle.

However, i believe what we discussed at the seminar and what you are touching on here a bit is the idea of not-correlated (). When one is going up, the other is going flat. In that case we are looking for numbers close to 0%, assuming to me that when one is trending the other is mostly flat.

Is that correct? I have a feeling the answer will be test it and find out.... :)

Hope all is well...

d-b
Full Member
Full Member
Posts: 20
Joined: Thu Apr 24, 2003 9:26 am

Post by d-b » Wed Dec 24, 2003 5:07 am

William,

The distinction you make is valid. According to your definitions, the same system trading two markets which are “un-correlatedâ€

William
Roundtable Knight
Roundtable Knight
Posts: 238
Joined: Sun May 04, 2003 4:41 pm
Location: Manhattan, New York

Thanks

Post by William » Wed Dec 24, 2003 9:05 am

Thanks David, in rereading my post i think i put too fine a point on the wording....thanks for the clarification.

Something i find slightly interesting is JWH has a trading system that trades the dollar vs. foreign currencies. Perhaps making a system that would focus in an area of strong uncorrelation may add some benefits if one were going to deploy a multi-system trading system.

http://www.jwh.com/templ007.cfm?id=007D ... &tid=007PP

BigBrad
Full Member
Full Member
Posts: 20
Joined: Tue Oct 28, 2003 8:39 pm

Post by BigBrad » Wed Dec 24, 2003 12:29 pm

Aren't you really determining two things with the correlations:

1. The closer to 0 the correlation between two series the less "predictive" they are, i.e. just because X went up it doesn't mean Y will go up. But at any point in time for short term they could be moving up together.

2. As the absolute value of correlation increase and approaches 1 then there is a "predictive" component. X is up so I expect Y to be up. If they are anti-correlated (-1) then is X is up I expect Y to be down.

Kiwi
Roundtable Knight
Roundtable Knight
Posts: 513
Joined: Wed Apr 16, 2003 1:18 am
Location: Nowhere near

Post by Kiwi » Thu Dec 25, 2003 8:11 pm

First -- Merry Xmas to everyone and I hope that your markets trend in your desired timeframes this year.

What follows is my opinion.

Correlation and negative correlation mean the same thing from a system development perspective. C=1 means the two commodities are identical with some factor of proportion. C= -1 means that when a goes down a certain amount then b goes up the same proportional amount.

What you want is a correlation close to zero. This means that although they might be correlated from time to time they are, on average, over the period you ran the correlation, not correlated. So, on average, over the period you ran the correlation, a does not alway go up when b goes up or vice versa.

Notice how many qualifications I put in there. YOU REALLY SHOULD. The reason is that uncorrelated commodities can become correlated for a short period of time ... and you will almost always perceive this at the worst possible time. When your stops get hit on currencies, gold, and energies at the same time.

So be careful of a nice general long term historical statistic like correlation.

And correlations are not predictive. A correlation is a historical statistic (like discovering that israeli airforce pilots where having girl babies ... then in the next period this was not true). What correlations can "show" is that the portfolio you chose because it shouldnt be correlated was uncorrelated in the period you just examined. Making assumptions about datamining is dangerous and very similar in its affects to overoptimisation.

Look for portfolios that are likely to have a low correlation ... but dont count on future = history if you want to be trading in 5 years time.

John

Post Reply