Measuring Correlation Between FX Pairs

What is the correlation between different FX pairs and how can it be measured?

As a Forex trader, it is necessary to be constantly aware of the characteristics of the currency market.  Without knowledge of the basic tendencies of any given currency pair, traders are exposing themselves to excessive amounts of risk.  One of the most beneficial pieces of knowledge for a Forex trader to be armed with is the historical correlation between currencies.  Understanding this relationship will not only allow traders to hedge positions, but it may also give them an edge when it comes to entering a trade. 

The term correlation is used to describe the relationship between two related variables and can be expressed as an integer between +1 and ?1.  A correlation of +1 is referred to as a perfect correlation and suggests that a 1-pip move in one currency should also see a 1 pip move in another currency in the same direction.  A ? 1 correlation implies that the two currencies move in the same magnitude in opposite directions.  If two variables have a correlation of zero, they move independent of one another and can be considered to have no relationship.

As you can imagine, knowing the exact relationship of one currency with another is priceless.  Many traders compare overlay charts in order to measure correlation, however it is far more precise to actually run a calculation in which you can get an exact figure.  Additionally, whether traders would like to admit it or not their judgment often contains a bias making it unwise to ?eyeball? a chart in order to determine correlation.  The simplest way to measure correlation is using Microsoft Excel?s Data Analyis Toolpak.

Microsoft Excel is capable of analyzing the correlation between two currency pairs at the click of a mouse.  The difficulty in the analysis lies within obtaining the data, rather than analyzing it.  If you have access to historical data, you can simply copy and paste the closing prices into respective columns in Microsoft Excel.  When I say closing price, it can be based on daily or intraday data as long as the beginning point and ending point are identical for the two currency pairs being compared.

After compiling the data into two separate columns, make sure that your Excel software is prepared for statistical analysis by clicking on the Tools tab and identifying the Data Analysis option on the drop down menu.  If you do not see Data Analysis as an option you must load the data analysis tool pak by choosing Add-Ins on the Tools menu and clicking on Analysis Toolpak and follow the instructions.

Once selected, the data is ready to be analyzed for correlation. Clicking on the Tools tab will reveal a drop down menu, which should now have Data Analysis as an option.  Clicking on Data Analysis will bring up a window in which you will be prompted to enter an ?input range?.  In other words, which data would you like to be considered in the analysis?  Make sure that you have chosen that columns rather than rows group the data.  Click and drag your mouse over the data that you would like to be involved in the calculation and press ?OK? in the window.  The result will be a table, providing the correlation coefficient to the relative currency pairs. 

The term correlation coefficient is used to represent the numerical representation of the relationship between the two variables.  As mentioned before, the range of the coefficient can be between +1 and ?1 with +1 representing a perfectly positive correlation and ?1 identifying a perfectly negative correlation.  In the example below, you can see that during the sample period the Swiss and the Euro have a positive correlation of .9667.  This is no surprise; it is a well-known fact that these particular currencies move in tandem.

If you understand concepts better when they are visualized, you could plot the data into a chart by clicking on the insert tab and selecting Chart.   A plotted chart best represents correlation.  If the chart reveals an inclining diagonal line, the two variables are positively correlated.  If the chart forms a declining diagonal line the variables are negatively correlated.  f the chart forms a declining diagonal line the variables are negatively correlated.  A random plot in which no particular pattern is formed indicates that there is no correlation between the variables.

Swiss/Dollar         Euro/Dollar

79.47                        121.53    
79.37                        121.58    
79.43                        121.33    
80.53                        122.68    
79.84                        121.71    
80.52                        122.28                               
80.51                        122.18                                          79.33                        120.25                             
79.21                        120.37    
78.88                        120.19    
78.82                        120.45    
79.09                        120.78    
80.01                        122.08    
80.13                        122.55    
78.55                        120.38    
77.67                        119.19    

                                   Swiss/Dollar            Euro/Dollar

Swiss/Dollar                       1 

Euro/Dollar                 0.966772227                   1      

To reiterate, correlation is simply a tool that can be used to improve trading results and should be seen as guidance rather than fool proof.  Although understanding the nature of the currency markets is imperative, you must also retain respect for the market. 

Carley Garner is the Senior Analyst at DeCarley Trading LLC where she also works as a broker.  Her book, "Commodity Options" published by FT Press will be on shelves in January 2009. Visit for your free subscription to Carley's e-newsletters and for details on the services she provides.  Carly is a Magna Cum Laude graduate of the University of Nevada Las Vegas, from which she earned dual bachelor’s degrees in both Finance and Accounting.Carley jumped into the options and futures industry with both feet and has become one of the most recognized names in the business.  Within months in the industry, she had published her first article in a nationally distributed periodical and hasn’t slowed down since.  She has been featured in the likes of Stocks and Commodities, Futures, Active Trader, Option Trader, Your Trading Edge, and Pitnews Magazine.  Carley is often interviewed by news services such as Reuters and Dow Jones Newswire, and has been quoted by the Investor’s Business Daily and the Wall Street Journal.  She has also been known to participate in Radio interviews. 

Carley Garner is the Senior Analyst at DeCarley Trading LLC where she also works as a broker.  Her book, "Commodity Options" published by F...


Well-known member
It is more usual to correlate returns i.e. ln(P[0]/P[1]) rather than the absolute price.
The author needs to go back to quant school.


Active member
I agree with GJ.

I would be interested in more resources that discuss correlation and its impact on risk across a portfolio of closely correlated markets.

Any one have any?



Well-known member
YachtFund said:
I agree with GJ.

I would be interested in more resources that discuss correlation and its impact on risk across a portfolio of closely correlated markets.

Any one have any?


"Modern Portfolio Theory & Investment Analysis" Elton & Gruber
"Active Portfolio Management" Grinold & Kahn

Both books cover this ground.
I've found that a "simple" correlation analysis, such as one based on the Pearson coefficient or the maximum of the correlation function (I don't know what Excel does but I'm guessing it's not too sophisticated), is often not particularly useful and can be misleading - particularly when you're interested in changes in correlation over time. The problem is that the data is essentially scaled so that each time series has unit (or whatever) variance. So if you take two fairly well correlated markets, the correlation coefficient is small when the markets are just going sideways (because the noise dominates), and close to unity when they're trending - nomatter what rates the two are trending with - because the trend dominates. So it can be useful for deciding if one pair of markets is more correlated than another pair, but actually not that useful for judging to what extent one particular market is tracking another for the purposes of seeing if one is becoming under/over valued with respect to the other. I'd be very grateful for any suggestions on more sophisticated correlation analysis methods.


The article is incorrect in stating that perfect correlation implies a change of the same magnitude in each underlying. If you correlate (for example) 2 sine waves which are in phase (i.e. they have the same frequency, their peaks and troughs happen at the same time) but with completely different amplitudes, they will be perfectly correlated. All that perfect correlation tells you is that a movement of X in one always produces a movement of Y in the other one, in the same direction (+1 correlation) or the opposite direction (-1).


Legendary member
Not exactly. A correlation coefficient indicates both direction and strength. You cite just +1 and -1. There is an enormous range between these two end points. :LOL:

The closer the coefficient is to either −1 or 1, the stronger is the correlation between the variables. But it's not just either +1 OR -1.


Experienced member
i think its all clap trap developed by academics to make them selves sound worth employing.

victor neiderhoffer - an academic who has made himself worth employing would agree as it happens.

ever changing cycles is the name of the game - whether its the dogs on a wednesday night at wimbledon track, or a fx cross.

as soon as everyone jumps on to x - be it the speed history of the dogs round a track or a correlation between 2 markets, the odds must change to reflect the increased competition/demand for price. the wise money is well on to this. so not only does the crowd move the odds out of their own favour through time, but also people waiting to 'copper' the crowd out of their bunce will come in to play. what used to work will no longer work. this is one of the only certainties we have in the market

this is why you can backtest an idea for ages and be 100% sure it works (in the past) yet as soon as you use it it stops working. chances are theres a load of other people who have spotted it too......

of course these things come back into fashion again at some stage, but the ever changing nature of cycles in life make sure that most will lose, irrelevant of how great the back testing seems to be, past correlations etc.


Established member
Maybe then we need to look for our correlations in ever-shorter time scales. Work out what statistical anomalies occurred yesterday and apply them today, for example?
charliechan said:
i think its all clap trap developed by academics to make them selves sound worth employing.

True, for this and also many other things. One of my roles as a quant used to be take various fancy-sounding mathematical techniques and technical indicators and show that they were indeed fairly useless for trading. Many of them were, at best, psychological crutches that the traders could use to support their intuition. For bond future spreads and other spreads, there were some common questions: is my spread likely to change much soon?, is one leg likely to lead the other?, which pair of futures would be best for trading as a spread? The timescales were typically 30 minutes to a few hours. In principle, some sort of correlation analysis could be good at answering these questions, but I never found anything more reliable than simply eyeballing the charts. I'm sure this "eyeballing" procedure could be coded up in some way, but such things tend to be considerably less reliable than an experienced trader.


Experienced member
GammaJammer said:
And thank f**k for that! or muppets like me wouldn't continue to get trading jobs, and I'm not sure I know how to do anything else anymore ;-)


lol - sounds like you have a head start on me! can i wire you some funds to look after?


Legendary member
GammaJammer said:
Course you can mate. I'll 'look after' them for you. ;-) Got a simple plan; Split the funds in half. With one half buy shares in the major brewing groups. Can you guess what I would do with the other half?
Just to show there is a definite advantage to correlation analysis I am going to suggest the other half would have a fairly close to +1 correlation to acquisition of late-night retail-end catering services & supplies.



Senior member
I'm sure this "eyeballing" procedure could be coded up in some way, but such things tend to be considerably less reliable than an experienced trader.

KISS principle wins.
I "eyeballed" both fundamental stats and charts for many years before being able to quantify what was the essence of the decision making process, now I have this process coded up and it is what I run.
Hardest thing with these algorithmic models is to stick to them of course not only in terms of level but also volume. Also hard to cut out everything else. Best article in latest "Trader" magazine was the interview with Larry Hite.
"to ensure strict adherence to the program, they even signed a written agreement to abide by their credos"

As for correlation. I always look at a correlation matrix of the products I trade and every month see how this has changed. I do not spend too much time on it more important to see that monthly returns remain overall positive. Correlation in futures seem to have increased across the board, not as easy to diversify these days but who as an active trader really cares so long as you are on the right side and you ensure your overall market exposure is not exceeding what you find acceptable.