How do you deal with currency correlation?

Hello fellow forum members,

I would like to ask how other traders here deal with currency correlation?
For example EUR/USD and USD/CHF have a -95% or greater correlation. So, if you are about to take a long on E/U, do you consider taking a short on the swissie? Or do you trade each pair individually, regardless of direction and regardless of the correlation?

Also, how short a time frame would the correlation be valid till? I've seen a reasonable correlation(eyeballing only) down to the 4hr TF. I havent tried looking for one on any lower tf's since i dont trade using them anymore.

Any opinions?

Having spent months studying correlation, even working on a system (with great help from Boxingislife) in an "attempt" to take advantage of it, my suggestion would be to trade each pair MOSTLY individually. You bring up the common example of EUR/USD and USD/CHF. Lets reverse the U/C to CHF/USD for reference purposes. It is true that they are consistent in their 80% - 95% correlation range. Having said that, that 5 - 20% difference can be entirely one sided in regards to EUR/CHF. The long term chart shows a constant trend in EUR/CHF despite their almost mirror performance against the USD on the short term. The chart on the link below references my point.

http://finance.yahoo.com/q/bc?t=1y&s...l&c=chfusd%3Dx

Although I am suggesting trading pairs individually, I would also suggest if you want to attempt to use correlation to your advantage ... try and use it to better entries and exits. One situation I commonly encounter is when I trade GBP/USD and EUR/USD. I know that these two pairs are often very closely correlated, because I know this it often benefits me to exit a EUR/USD trade when I get a signal to do so on GBP/USD. Or establish a position in GBP/USD when I get a signal on EUR/USD etc etc. Although it is very common for one pair to outperform the other in a certain direction ... it is quite rare that one will be heading long with authority, while the other is heading short. In regards to my example, even if I did exit both pairs while only getting a GBP/USD signal, if the signal proves to be correct and the GBP/USD would have gone against me ... it is unlikely that the EUR/USD would really be going well for me. Does that make sense? The EUR/USD may still be flat because it is outperforming the falling GBP/USD, though my exit will less commonly prove to be bad in regards to my system.

If you are greatly interested in correlations, I would sugget using the demo from www.fxsol.com and use their AccuCharts. You can overlay and compare a lot of different chart views in regards to correlation.


Could you expand more upon this? I do believe I know what you are speaking about, I am curious what your perspective is on the LONG TERM correlation. Extremes do work, especially with historically correlated pairs such as GBP, CHF, EUR ... however, what would your criteria be for extreme and how would you manage long term fluctuations as occurs steadily over 5 year and plus periods?

I just look at a 4 hour layered chart. Its pretty self explanitory.

http://i131.photobucket.com/albums/p.../Picture-2.gif

Thanks for the replies so far...

The problem I have in trading correlated pairs is that if the position(s) goes against you, then you will have essentially increased your risk!

StockKJay, I do tend to use something similar to what you decribed though. The volatility in the cable, makes it easier to see where the E/U might be going...

xPip, could you please explain what you meant?

Good analysis, SJ.

Just wantd to add that relying on "regressing ot the mean" is really a rather dangerous concept in FX (or any other human activity). Check out the examples given in the Wikipedia article:
http://en.wikipedia.org/wiki/Regression_toward_the_mean
(especially the bit about Horace Secrist, under "Regression Fallacies"). FWIW, it's also very easy to construct probabilities that have no mean (and hence nothing to return to). Not even if you waited till entropy makes the universe run down (& think of the roll-over cost), while I'm sure somewhere on the planet there is probably at least one trader waiting for U/J to get back to the 300 level (and lots more waiting for Usd/Cad to get to 1.50 or so...)

As a bit of a diversion there is a correlation that is forced in FX that doesn't exist in stocks, so it makes pair correlation much harder to use as a diversity measure. To see this, look at E/U, U/J, and E/J . If E/J goes up, then by necessity the product E/U.U/J must also go up, otherwise big arb opportunity. Ditto for every set of pairs E/X.X/J, where X is any country with a tradable ccy (NZD comes to mind). If you do your correlations in log space (I've never bothered) you might be able to get something usable that would give you an estimate of the expected spike in other ccy's given an unpredicted spike (like an intervention) in one pair. (I don't bother because I assume folks like Rentec or BoJ already have much faster models to arbitrage that sort of thing automagically...)