Sector Limits

[LeviF]

For those of you who use sector limits, did you perform any correlation analysis to group the instruments, or did you just arbitrarily use the standard metals, meats, grains, etc groups?

[AK1]

Hi to all,

This is my 1st post on Trading Blox...
Ive learnt alot from reading post..

This is something which has bothered me.
Many books talk about limits based on sectors (metals, grains etc), the issue with this approach is that it does not prevent you from have a number of highly correlated trades (eggs in basket).
For this reason - isnt grouping based on correlation more logical ?

[sluggo]

For those who trade global diversified portfolios containing 60 or 120 (or more!) futures markets, this seems to be more of a problem in theory than in practice. Agricultural futures in Malaysia, for example, do not whip around a lot when the Fed holds an FOMC meeting. Greasy Wool isn't much affected by Quadruple Witching Hour (the 3rd Friday of March, June, Sept, Dec, when quarterly series stock index options and futures expire). Times "when a Big Number comes out" (unemployment figures, inflation figures, OPEC production targets, etc.) tend not to spike the price of Lead. And so forth. If you've got lots of positions, in lots of different "things" all around the world, a shocking big move in X only affects some of them, not all of them. Furthermore if you've got lots of positions, each position is necessarily small. Lessening your exposure to a price shock like the ~~ 12 ATR price gap shown on p.93 of WOTT.

The large portfolio, globally diversified futures traders that I'm friends with, generally do one big correlation study on historical data. Then they group markets into "similarity classes" (call them sectors if you like) which they don't adjust thereafter. There isn't any ongoing correlation analysis; the similarity classes don't change from week to week or from year to year. These traders figure out their groups or sectors or classes, once, and then leave them alone.

Often there will be surprises in the groupings (for example, where would you put "Goldman Sachs Commodity Index futures"? Do you subdivide the interest rate futures, and if so, how? By geography? By maturity? By volatility?), but a large number of futures wind up exactly where you'd expect: Gold, Silver, Platinum, Palladium are all members of the same class; Corn, Wheat, Soybeans are all members of the same class (different from Gold's!), etc.

After the similarity classes are chosen, in day to day trading they are used to avoid unbalanced or "over weighted" situations in which one class contributes a lot more risk or volatility or margin requirement or VaR or simple position count, than its fair share. You do a "Resizing", an adjustment of positionsize in the middle of a trade.

EDIT - added reference to Way Of The Turtle (book)

[LeviF]

The large portfolio, globally diversified futures traders that I'm friends with...

Where does one find these friends? Trading seems to be a lonely business.

[sluggo]

viewtopic.php?p=33247&highlight=hermi%2A#33247

[AK1]

Thanks Sluggo,

You make a very good point. Anyone with the ability to trade such a large and diversified portfolio benefits from having a large number of contracts that will not all react to X event in the same way. However, this may not be the case for traders with a basket of 20-40 products.
I am assuming that with a smaller basket, this concept become more important.

In WOTT page 256, they mention a set of rules on on "single market, closely, loosely, single direction" trades. I was trying to get my head around how one defines a closely or loosely correlated market;
copper and gold for instance (loosely)? This seems subjective.. Or do we let correlation stats dictate what a closely or loosely correlated trade is. I guess the issue is that correlation relationships can change... while sectors cant... gold with always be a metal. What are you thoughts?
Its clear that large drawdowns can be a result of having too many correlated positions move against you at the same time.. and sector or correlation limits seems to provide a viable way of trackling this

Another unrelated question: When backtesting, do most of you only use $ deminated contracts? Or does TB enable you to backtest a across a different fx demominated contracts? What a convential way of tackling this issue ?

[sluggo]

The Turtle experiment ended 21 years ago. Yet some people remain intrigued and fascinated by it, even today. They thirst for details, facts, and specifics, even after they study pictures such as this one: (Turtle picture)

An excellent source of Turtle facts is the Original Turtle Trading Rules document in Acrobat .pdf format, that you can find floating around on various internet sites. I've attached an image below, showing what a Google "search hit" looks like when you've found it. Carefully note: it is a pdf document and the Google result begins with "[PDF]". A quick doublecheck is to open the .pdf document and view the first page; if you see the theree orange-and-purple turtles pictured below, you've got the right thing. This .pdf document was written by an Original Turtle.

Another excellent source is the book WOTT, written by the same Original Turtle.

A third excellent source is the Turtle Trading System software code, written by an Original Turtle, and sold by our hosts on this Roundtable Forum: Trading Blox. Like many here, I myself am a satisfied customer of this software. Very satisfied.

I've attached the top few and bottom few entries of Trading Blox's "Futures Dictionary", pertaining to the Turtle Trading System's definitions of Closely Correlated and Loosely Correlated futures markets. There are some small surprises in it, such as (1) the fact that Wheat is not considered "Closely" correlated to anything, and (2) the fact that 10 Year Notes are "Loosely" correlated to Gold, but 30 Year Bonds are not.

Someone who is absolutely consumed by thirst for Turtle facts and details, will naturally want to see the remainder of the table immediately: Stat! One way to get it, and the one I happen to recommend, is to buy the software.

Final thought: the Turtle experiment ended 21 years ago.

[LeviF]

I wonder if these correlations were the same as those defined back in the 80's? There is no mention in any of the literature as to what is really "closely" and "loosely" correlated.

[RedRock]

The Turtle experiment ended 21 years ago. Yet some people remain intrigued and fascinated by it, even today. They thirst for details, facts, and specifics, ...

Final thought: the Turtle experiment ended 21 years ago.

Some thoughts from the horses mouth:

http://www.curtisfaith.com/2009/10/12/q ... s-markets/

[RedRock]

Its clear that large drawdowns can be a result of having too many correlated positions move against you at the same time.. and sector or correlation limits seems to provide a viable way of trackling this

I wonder if these correlations were the same as those defined back in the 80's? There is no mention in any of the literature as to what is really "closely" and "loosely" correlated.

Ralph Vince, in his book The Handbook of Portfolio Mathematics, suggests that correlation studies in a traditional sense are meaningless. As many here already know, when a large sigma event occurs, all anglo markets become correlated for the period. Sluggos' 'spanning the globe' suggestion is really the only way out of this other than dealing with the reality and keeping bets small. Its the temporal correlation rather than historic, which is of concern.

leviejean, I think that as you allude to, trendfollowing has grown in popularity, so has the correlation of formerly unlike sectors, as all the elephants try getting through the same doors at once.

[LeviF]

The Turtle experiment ended 21 years ago. Yet some people remain intrigued and fascinated by it, even today. They thirst for details, facts, and specifics, ...

Final thought: the Turtle experiment ended 21 years ago.

Some thoughts from the horses mouth:

http://www.curtisfaith.com/2009/10/12/q ... s-markets/

I did not know this site existed. Thanks for the link.

[kianti]

Any dendogrammer around?

The complexity in dealing with correlated market lies in correctly identifying correlated groups. With just two markets this is simple enough, the market is either correlated or not.
With three things get a little more complicated. Imagine three markets A,B & C, if A is highly correlated with C and B is highly correlated with C then a naïve observer would probably suggest that A and B were highly correlated, unfortunately this isn’t necessarily true, hence the rise in complexity. It should be clear at this stage that if you increase the
number of markets in a portfolio to 60 to 100 (usual levels for managed futures funds) then working out all the correlated sets in the data isn’t straightforward.
A problem of this type can be tackled using dendrograms or graphing theory. An example dendrogram is shown below. The “Distanceâ€

[sluggo]

Another way to approach it is to search for structure within the correlation matrix. (An N x N matrix whose entry C[j][k] is the correlation coefficient between market #j and market #k). As always, I recommend calculating the correlation of daily returns rather than daily prices).

For example the Data Mining crowd loves to hunt for structure by performing "Cluster Analysis", partly because they have efficient algorithms they can apply to very large problems: Partition the set of markets into clusters, such that the markets within a cluster have large correlations to each other, and small correlations to markets in other clusters. When you've found the clusters, draw a picture. If you want to, call that picture a Dendrite Mammogram or something.