How to deal with correlated instruments
-
- Roundtable Knight
- Posts: 1436
- Joined: Mon Dec 22, 2003 12:24 pm
- Location: Des Moines, IA
- Contact:
How to deal with correlated instruments
My forex broker has just introduced 8 new currency pairs. The new pairs are simply derivatives of the already offered existing pairs. For example, instead of only offering NZDUSD, NZDJPY and EURNZD are now offered. Some of these new pairs (and many of the existing pairs) by nature are highly correlated with other pairs. It seems to be overly risky to me to have a position in a pair like NZDUSD and also in NZDJPY. However, if I arbitrarily selected just one of those pairs to have in my portfolio, I would hate to miss a good trade in the other.
My broker has access to 26 individual currencies. Over time, if they introduced every possible pair (extreme example), that's 650 instruments, of which a very substantial portion would be relatively highly correlated with other instruments.
Any suggestions on how to deal with this?
My broker has access to 26 individual currencies. Over time, if they introduced every possible pair (extreme example), that's 650 instruments, of which a very substantial portion would be relatively highly correlated with other instruments.
Any suggestions on how to deal with this?
I would look at it two ways, either alone or in combination:
1) Use Seykota's concept of "Portfolio Heat," (which is total open risk for all trades) and apply it to groups of currencies.
Lets say you take 1% risk on every trade. You can limit yourself to 2.5% "heat" on all common-currency-common-direction pairs. So, if you get a long signal on NZDEUR, NZDUSD and NZDJPY, you only take 0.83% risk on each trade, since 1% on each would violate the 2.5% heat restriction.
2) Manage exposure to each currency individually. There is a screen on Oanda (a forex broker) which shows the net long/short USD exposure to each individual currency. You can then "eyeball" the exposures to make sure no one currency is out of balance. This would be more of a sanity check than anything else.
1) Use Seykota's concept of "Portfolio Heat," (which is total open risk for all trades) and apply it to groups of currencies.
Lets say you take 1% risk on every trade. You can limit yourself to 2.5% "heat" on all common-currency-common-direction pairs. So, if you get a long signal on NZDEUR, NZDUSD and NZDJPY, you only take 0.83% risk on each trade, since 1% on each would violate the 2.5% heat restriction.
2) Manage exposure to each currency individually. There is a screen on Oanda (a forex broker) which shows the net long/short USD exposure to each individual currency. You can then "eyeball" the exposures to make sure no one currency is out of balance. This would be more of a sanity check than anything else.
Reorganizing my bookshelves, I notice that I've accidentally omitted two other Forex authors who are just as famous as those four: Kathy Lien and Boris Schlossberg. There was a thread about BS a while ago, at viewtopic.php?p=21596&highlight=schloss%2A#21596sluggo wrote:What do the book authors and Forex Educators say? Cornelius Luca, Raghee Horner, Paul Bishop, Julian Walmsley, people like that?
I was thinking about doing some research about this and I bet some of you have already done it and, perhaps, I could spare my time and money (since I don't have good historical data of cross-rates).
Premise:
If the base currency is, for instance, USD then it doesn't make sense trading cross-rates. An example might clarify. Let's make a case:
1) System A is long EURUSD which is in an uptrend; and short USDJPY which is in a downtrend;
2) EURJPY = EURUSD/JPYUSD, so
3) if System A is also long EURJPY then it's basically long EURUSD and long USDJPY (short JPYUSD).
So I suspect that trading cross-rates ends up in a big mess and is counterproductive.
Any ideas?
Premise:
If the base currency is, for instance, USD then it doesn't make sense trading cross-rates. An example might clarify. Let's make a case:
1) System A is long EURUSD which is in an uptrend; and short USDJPY which is in a downtrend;
2) EURJPY = EURUSD/JPYUSD, so
3) if System A is also long EURJPY then it's basically long EURUSD and long USDJPY (short JPYUSD).
So I suspect that trading cross-rates ends up in a big mess and is counterproductive.
Any ideas?
-
- Roundtable Knight
- Posts: 1436
- Joined: Mon Dec 22, 2003 12:24 pm
- Location: Des Moines, IA
- Contact:
I just ran two tests of my forex system.
First is on all 40 instruments in my portfolio, including 21 foreign crosses. R-cubed of 2.23.
Second is on my 19 USD-only pairs. R-cubed of 1.26.
Based on this limited testing, your premise is incorrect.
But as always, I wouldn't take mine or anyone else's word for it.
First is on all 40 instruments in my portfolio, including 21 foreign crosses. R-cubed of 2.23.
Second is on my 19 USD-only pairs. R-cubed of 1.26.
Based on this limited testing, your premise is incorrect.
But as always, I wouldn't take mine or anyone else's word for it.
Thanks. I just made a small test against the correlated groups.
For instance, run a test against EURUSD, GBPUSD, AUDUSD, USDJPY, USDCHF and USDCAD. And then, afterwards, include the cross-rates of all these currencies: EURJPY, EURAUD, EURCAD, EURGBP, EURCHF, GBPAUD, GBPJPY, GBPCHF, GBPCAD, AUDJPY, AUDCHF, AUDCAD, CHFJPY, CADJPY, CADCHF.
I used the following method, for this test:
1) I optimized two basic trend following systems against the six major currency pairs, with EOD data that goes back to 14 Jan 1987.
2) I tested the system against the six major currency pairs from Jan 1999 to December 2008, using the free historical data of TradingBlox.com.
3) I made the corresponding adjustments in account equity and position sizing and tested the system against the 15 currency pairs from Jan 1999 to December 2008, using the free historical data of TradingBlox.com.
The discrepancy between results is disheartening, going from a MAR ratio of 0.81 in the six majors to a MAR ratio of 0.07 in all of the sample.
Perhaps, there is a flaw in step 1, i.e., the systems were tested against the majors only. Whatever...
For instance, run a test against EURUSD, GBPUSD, AUDUSD, USDJPY, USDCHF and USDCAD. And then, afterwards, include the cross-rates of all these currencies: EURJPY, EURAUD, EURCAD, EURGBP, EURCHF, GBPAUD, GBPJPY, GBPCHF, GBPCAD, AUDJPY, AUDCHF, AUDCAD, CHFJPY, CADJPY, CADCHF.
I used the following method, for this test:
1) I optimized two basic trend following systems against the six major currency pairs, with EOD data that goes back to 14 Jan 1987.
2) I tested the system against the six major currency pairs from Jan 1999 to December 2008, using the free historical data of TradingBlox.com.
3) I made the corresponding adjustments in account equity and position sizing and tested the system against the 15 currency pairs from Jan 1999 to December 2008, using the free historical data of TradingBlox.com.
The discrepancy between results is disheartening, going from a MAR ratio of 0.81 in the six majors to a MAR ratio of 0.07 in all of the sample.
Perhaps, there is a flaw in step 1, i.e., the systems were tested against the majors only. Whatever...
-
- Roundtable Knight
- Posts: 138
- Joined: Wed Nov 10, 2004 4:36 pm
Alp,
I would argue that, conceptually, trading forex cross rates is akin to spread trading in commodities and as such, why would you expect a LTTF approach with LTTF parameters based on the majors to work in such a spread trading scenario? This might explain why there was such a reduction in portfolio performance.
I would argue that, conceptually, trading forex cross rates is akin to spread trading in commodities and as such, why would you expect a LTTF approach with LTTF parameters based on the majors to work in such a spread trading scenario? This might explain why there was such a reduction in portfolio performance.
Exactly. As in the example I posted before, EURJPY = EURUSD/JPYUSD.babelproofreader wrote:Alp,
I would argue that, conceptually, trading forex cross rates is akin to spread trading in commodities and as such, why would you expect a LTTF approach with LTTF parameters based on the majors to work in such a spread trading scenario? This might explain why there was such a reduction in portfolio performance.
Let's compare some rates:
Code: Select all
05/10/2000:
EURUSD: 0.9060
USDJPY: 109.50 JPYUSD = 1/USDJPY = 0.00913242009
EURJPY: 99.28 EURUSD/JPYUSD = 99.207
12/31/2008:
EURUSD: 1.3967
USDJPY: 90.64 JPYUSD = 1/USDJPY = 0.01103265666
EURJPY: 126.63 EURUSD/JPYUSD = 126.60
If EUR is uptrending, conversely, USD is downtrending, so a trend following robust approach will likely be long EURUSD and short USDJPY (or long JPYUSD) simultaneously.
So the cross-rate only represents the spread between EUR and JPY. The problem, as I see it, is that most currencies are still pegged against the dollar. As such, trading cross-rates is much like arbitrage playing.
-
- Roundtable Knight
- Posts: 1436
- Joined: Mon Dec 22, 2003 12:24 pm
- Location: Des Moines, IA
- Contact:
Yes. I think I need to implement exposure limits to each currency. However, coming up with limits that make sense and are consistent will be difficult. For example, I can probably afford to take on larger positions with USD, EUR, JPY than HUF, MXN, & TRY, but I wonder how to objectively determine that...
-
- Roundtable Knight
- Posts: 1436
- Joined: Mon Dec 22, 2003 12:24 pm
- Location: Des Moines, IA
- Contact:
I found some sizing constraints from a currency fund that manages $3B. If its good enough for them, its probably good enough for me.
Code: Select all
- 150% of equity: EUR, JPY
- 70% of equity: AUD, CAD, CHF, CZK, GBP, HUF, KRW, PLN, SEK
- 60% of equity: BRL, ZAR, MXN
- 50% of equity: NOK, SGD, NZD
- 35% of equity: CLP, CNY, ILS
- 25% of equity: IDR, INR, PHP, MYR, TRY, TWD
- 10% of equity: ARS, COP, ISK, PEN, RON, RUB