portfolio question

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
tigertaco
Full Member
Full Member
Posts: 23
Joined: Fri Feb 27, 2009 1:50 am

portfolio question

Postby tigertaco » Mon Aug 10, 2009 1:15 am

I would like to build a list of 25 or so low correlated markets to test my systems. Does anybody have a suggestion on how to go about creating such a list? I want it to include stocks, forex, and futures traded at different exchanges.

I would really appreciate if you give me an example of a portfolio that you are using (or a part of it).

thanks, tt

User avatar
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2969
Joined: Fri Jun 11, 2004 2:50 pm

Postby sluggo » Mon Aug 10, 2009 8:35 am

If you decide to measure correlation yourself, I suggest you do so upon price returns rather than prices. I seem to remember a few threads about the topic here on the roundtable forum in days gone by, perhaps Search might find something useful. Handy tip: Microsoft Excel includes a built in function "CORREL()" that calculates correlations.

Data vendor CSI (link) sells an extra cost service which measures correlation. They wrote two articles / sales pieces about the service for their customer newsletter; these are on their website too. I uploaded the first three pages of one of them, attached below. Finally, CSI also offers a "free portfolio rating" which may be useful (or may not): http://www.csidata.com/cgi-bin/FreePort ... lancing.pl .

A project like this one opens up all kinds of uncomfortable questions that are hard to answer initially. Such as "Am I interested in correlations of daily price returns? Weekly? Monthly? Am I interested in correlations over the past 1 year? 3 years? 6 years? All the price history I've got? What do I mean when I say "Low" Correlation? How low is low? After I use correlation to help me choose a portfolio, how often will I update the portfolio? How often will I perform this entire study, again and again?"

A somewhat more advanced question is: "Am I interested in the correlations of instrument price returns (which are the INPUTS to mechanical trading systems), or am I interested in the correlations of instrument equity returns (which are the OUTPUTS of mechanical trading systems)?" In my experiments, with the kinds of systems and instruments I typically test, the outputs of trading systems quite often have lower correlation than the inputs. But of course that's just anecdotal; try it for yourself and see.
Attachments
part_001.pdf
Forum attachments must be less than 1000 Kbytes; here are 800 Kbytes from CSI
(801.99 KiB) Downloaded 439 times

User avatar
LeviF
Roundtable Knight
Roundtable Knight
Posts: 1425
Joined: Mon Dec 22, 2003 12:24 pm
Location: Des Moines, IA
Contact:

Postby LeviF » Mon Aug 10, 2009 9:42 am

sluggo wrote:A project like this one opens up all kinds of uncomfortable questions that are hard to answer initially.

I have spent a lot of time on this subject and have not yet reached any conclusions on the answers to these "uncomfortable questions". The problem you run into is correlations change so much from period to period that relying on past measurements to make decisions for the future is very difficult.

User avatar
RedRock
Roundtable Knight
Roundtable Knight
Posts: 939
Joined: Fri Jan 30, 2004 3:54 pm
Location: Chicago

Postby RedRock » Mon Aug 10, 2009 1:33 pm

levijean wrote:
sluggo wrote:A project like this one opens up all kinds of uncomfortable questions that are hard to answer initially.

I have spent a lot of time on this subject and have not yet reached any conclusions on the answers to these "uncomfortable questions". The problem you run into is correlations change so much from period to period that relying on past measurements to make decisions for the future is very difficult.


Yup. Its THE perfect non correlated portfolio... Until that morning you turn on the computer to discover... today its perfectly correlated. :shock:

User avatar
LeviF
Roundtable Knight
Roundtable Knight
Posts: 1425
Joined: Mon Dec 22, 2003 12:24 pm
Location: Des Moines, IA
Contact:

Postby LeviF » Wed Aug 26, 2009 12:24 pm

sluggo wrote:A somewhat more advanced question is: "Am I interested in the correlations of instrument price returns (which are the INPUTS to mechanical trading systems), or am I interested in the correlations of instrument equity returns (which are the OUTPUTS of mechanical trading systems)?" In my experiments, with the kinds of systems and instruments I typically test, the outputs of trading systems quite often have lower correlation than the inputs. But of course that's just anecdotal; try it for yourself and see.

I'm having trouble comprehending why the outputs would have a different correlation than the inputs (assuming the same time frame, system, etc).

User avatar
RedRock
Roundtable Knight
Roundtable Knight
Posts: 939
Joined: Fri Jan 30, 2004 3:54 pm
Location: Chicago

Postby RedRock » Wed Aug 26, 2009 12:55 pm

different systems

User avatar
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2969
Joined: Fri Jun 11, 2004 2:50 pm

Postby sluggo » Wed Aug 26, 2009 1:18 pm

levijean wrote:I'm having trouble comprehending why the outputs would have a different correlation than the inputs (assuming the same time frame, system, etc).

What does the data say? What is the outcome of this sequence of experiments?

(1) Measure the correlation of ES price returns to YM price returns
(2) Run a trading sytem on ES prices. Capture the equity curve
(3) Run the exact same trading system on YM prices. Capture the equity curve.
(4) Measure the correlation of ES equity curve returns to YM equity curve returns
(5) Which is greater, the correlation in (1) or the correlation in (4)?
(6) Repeat the above using ES and SP rather than ES and YM.
(7) Repeat many times using many pairs of instruments
( 8 ) Draw a conclusion

User avatar
RedRock
Roundtable Knight
Roundtable Knight
Posts: 939
Joined: Fri Jan 30, 2004 3:54 pm
Location: Chicago

Postby RedRock » Wed Aug 26, 2009 2:30 pm

Ah, I see. Ive noticed this too. That sometimes pairs or groups of what should be very similar markets. Can play very nicely together indeed. One gets an entry the others dont reach. One gets stopped out but the other position remains. etc etc etc Of course the flip side is that sometimes they all get entries, and the next day all get stopped out. So some moderation may yet be advised...

User avatar
svquant
Roundtable Knight
Roundtable Knight
Posts: 124
Joined: Mon Nov 07, 2005 3:39 am
Location: Silicon Valley, CA
Contact:

Postby svquant » Wed Aug 26, 2009 6:14 pm

I think I've posted on this once before in response to a Sluggo inspired thread...

Do your own experiments and be comfortable with the results and understand them. In general there appears to be a band of correlation between price series where the correlation between trading systems is even less. Perhaps the trading system is further "whitening" the series creating even less correlation for these low (to perhaps medium) correlated instruments. When there is high correlation, either positive or negative, the noise added by the trading system will not change the correlation and for highly negative correlated instruments could actually create positive correlation - which may not be what you are looking for.

For example, if two series have a -1 correlation on returns, they can easily have a +1 correlation on equity curves (or returns) out of a trading system. Think of how a trend following system that goes long/short would trade an instrument that had a nice trend of +10%/year and one that had a trend of -10%/year (i.e. an exact inverse of the +10% time series)

Your mileage may vary, past correlation is no guarantee of future correlation, there is a risk of loss when trading instruments that are correlated, and a risk of loss in trading instruments that are not correlated (hey that is what disclosure docs basically say :? )

Roscoe
Roundtable Knight
Roundtable Knight
Posts: 246
Joined: Sat Jan 24, 2004 2:06 am
Location: Houston TX
Contact:

Postby Roscoe » Wed Aug 26, 2009 7:08 pm

My $0.02 worth: correlation is essentially meaningless for all practical purposes. And when there is a major move all markets become correlated anyway. I can see (and find) no way to make use of correlation information.

User avatar
Chris67
Roundtable Knight
Roundtable Knight
Posts: 1035
Joined: Tue Dec 16, 2003 2:12 pm
Location: London

Postby Chris67 » Thu Aug 27, 2009 3:21 am

concur with the above - correlation is meaningless - when the preverbial s$%^& hits the fan everything becomes correlated

One thing to assume is that if cnbc and all economists and analysts are currently to be believed - the fact that the Western Economies stand on the brink of total enihalation will have no effect on China or other far eastern markets - which are all merrily going to lead the way out of recession for the global economy !! so theres some un-correlated markets

Also my favourite non- correlated market is Economic Reality / Common Sense / Truth is totally un-correlated to markets/ what comes out of governments and central banks mouths - this pair has never been correlated and never will be

User avatar
RedRock
Roundtable Knight
Roundtable Knight
Posts: 939
Joined: Fri Jan 30, 2004 3:54 pm
Location: Chicago

Postby RedRock » Thu Aug 27, 2009 11:06 am

but which is less worse? Putting a full size position on in say, the beans. Or first putting a much smaller position in beans, followed by a similar fractional position in corn, joined the next day, by a fractional position in wheat.

It muddies the water somewhat with risk management. But what will yield a smoother ride? The full size beans, or the third sized beans,corn,wheat.

The above may be extrapolated across the spectrum of tradables.

Perhaps another topic entirely. But how about assembling a synthetic grain index. composed proportionally of all the grains, for example. Then when a signal is given for the index, a fractional position is placed in each component.

Just some thoughts as yet untested...

User avatar
sluggo
Roundtable Knight
Roundtable Knight
Posts: 2969
Joined: Fri Jun 11, 2004 2:50 pm

Postby sluggo » Thu Aug 27, 2009 11:17 am

A fun idea to add to your testing queue: suppose you have decided that (Corn, Beans, Wheat) move together. Suppose you get a signal to enter Corn. On the theory that (Corn, Beans, Wheat) move together, perhaps you may decide to enter all three when you get the first signal*. You enter "on time" in the first, and "early" (therefore, more profitably??) in the other two.

Or maybe you despise being early; maybe you feel that trying to enter early means lots of false moves and stopouts for a quick loss. In that case, perhaps you may want to wait until you have entry signals for all three, before you actually take a position in any of them. When you're flat all three and you get an entry signal in Corn: do nothing. Then when you get an entry signal in Wheat: do nothing. Finally when you get an entry signal in Beans: take all three signals. Confirmation, confirmation, confirmation.

Or maybe you think "the earliest signal is usually the strongest move". In that case, when you're flat all three and you get a signal to enter Corn, you enter Corn. When you get a signal to enter Wheat, you don't enter Wheat, instead you enter another position in Corn (because Corn was earliest, and earliest is bestest). When you get a signal to enter Beans, you enter a third position in Corn. Woo hoo, what fun.

* Call this "one out of three". You can modify it to become "two out of three" if you like.


Return to “Testing and Simulation”

Who is online

Users browsing this forum: No registered users and 2 guests