Back to Guides
Guide

Which Forex Pairs Are Most Correlated, and By How Much?

Which forex pairs are most correlated: the pairs every source names, the figures they publish side by side, and why a coefficient describes a window.

EUR/USD and GBP/USD are the pair every source names for the strongest positive correlation in forex, and EUR/USD against USD/CHF for the strongest negative one. The pairs are settled, but the numbers are not: published figures for that first pair run from 0.56 to 0.95.

A coefficient is a property of the window you measured over, not of the two symbols, and most correlation tables leave the window out. Change the dates, the timeframe or the hour of day and the number changes with them, so a figure with no window attached cannot be checked and cannot be used.

Our own historical data comes from Dukascopy, which also publishes one of the tables quoted here. It is treated like the others.

Which forex pairs are most correlated?

EUR/USD and GBP/USD, on every source. What each publishes for that pair and the strongest negative one, and whether it states a period:

SourceEUR/USD vs GBP/USDEUR/USD vs USD/CHFStates its window?
dukascopy.com"approximately +0.95""around -0.95"No
ig.com0.81-0.54 in the matrix, -0.967 to -0.933 in a second table from the day beforeYes: "calculated over a one-day period on 26 November 2019 using the Pearson correlation coefficient"
newyorkcityservers.com"+0.80 and +0.95 on daily charts""often between -0.85 and -0.98"Timeframe only, no period
blueberrymarkets.com"0.81 and 0.95 (81% to 95%)""-0.85 to -1 (-85% to -100%)"No
logikfx.com89%-91%Only "at the time of writing this post"
benzinga.com0.56, credited to myfxbook.com-0.81"recent daily correlation coefficients", though its grid prints 0.00 where a pair meets itself, so read it with care
blog.opofinance.comNamed first, no figureNamed third, no figureNo coefficients at all

The published figures for one pair span 0.56 to 0.95, and the ranking is not settled either, even among the sources that publish numbers. Dukascopy puts AUD/USD and NZD/USD at about +0.95, matching its EUR/USD figure. newyorkcityservers.com has the Australasian pair at "+0.85 and +0.95" against "+0.80 and +0.95" and calls it "one of the strongest in the market". blueberrymarkets.com has it at "0.86 and 0.99" against "0.81 and 0.95". On their own numbers the most correlated pair is arguably AUD/USD and NZD/USD.

What is correlation in forex?

The statistical relationship between two pairs: whether they move in the same direction, in opposite directions, or independently. IG states the ends plainly, that the coefficient "ranges from 1 to -1, with 1 representing a perfect positive correlation and -1 representing a perfect negative correlation", with 0 meaning no correlation.

The same measure gets printed on three scales. Most sources use -1 to +1, blueberrymarkets.com states the range as "-100 to 100", and logikfx.com prints percentages, "EUR/USD vs GBP/USD = 89% correlated". Dukascopy uses both on one page, writing that "Two markets will always move in the same way if they are 100% correlated". 0.89 and 89% are the same reading, but compare a decimal table against a percentage table without noticing and you misread one by a factor of a hundred.

Is there a correlated forex pairs list?

Yes. Between them, these sources name the following pairs, and none of the coefficients is ours.

Positive: EUR/USD and GBP/USD; AUD/USD and NZD/USD; EUR/JPY and GBP/JPY; EUR/USD and AUD/USD, published at "+0.60 to +0.80" by newyorkcityservers.com, 0.79 in IG's one-day matrix and 56% by logikfx; EUR/JPY and USD/JPY; GBP/USD and EUR/JPY; USD/CHF and GBP/CHF.

Negative: EUR/USD and USD/CHF; EUR/USD and USD/CAD; AUD/USD and USD/CAD; GBP/USD and USD/JPY; AUD/USD and USD/CHF; GBP/USD and USD/CAD.

Dukascopy's FAQ answers "What forex pairs are not correlated?" with "The least correlated forex pairs are EUR/JPY and USD/CAD", while IG measured that same pair at -0.71 over its one-day window, and by newyorkcityservers.com's bands anything past -0.70 is a strong negative correlation.

A list like this tells you which pairs to check, and your own measurement over your own dates tells you what they did.

Why do correlation tables disagree?

Because a coefficient describes a window, and almost nobody prints the window.

IG publishes -0.54 for EUR/USD against USD/CHF in a matrix "calculated over a one-day period on 26 November 2019". On the same page it publishes an hour by hour table for the trading day before, running from -0.967 at 8am to -0.933 at 9pm UK time, with -0.970 as its strongest reading. The two readings are the same pairs from the same source one day apart, and they differ by more than 0.4. Its body text offers a third figure: "usually below -0.70, but it can go as low as -0.97."

newyorkcityservers.com states the rule in one sentence: "correlation is time-frame dependent. A pair showing +0.90 correlation on daily charts might only show +0.55 on 1-hour charts", and its advice follows, "Always match your correlation analysis to the time frame you're actually trading."

Two other reasons. A ranking can have nothing behind it: blog.opofinance.com orders seven pairs by strength without publishing a coefficient for any of them. And a coefficient gets overstated: blueberrymarkets.com says 0.81 to 0.95 implies "that any increase in EUR/USD will definitely lead to an increase in GBP/USD exchange rates as well and vice versa". A reading of 0.81 makes nothing definite, and IG puts it the other way: "perfectly correlated currency pairs are very rare, and there is always a degree of uncertainty when trading the financial markets."

What is a correlation matrix, and how do you spot a broken one?

A grid with the same pairs down the side and across the top, where each cell holds the coefficient for the two that meet there. newyorkcityservers.com calls it "the most efficient way to see all your correlation relationships at a glance".

Three checks. The diagonal must read 1, because every series is perfectly correlated with itself, so where a pair meets itself the cell has to be 1.00, or 100% in a percentage grid. Zeros or blanks there mean the grid was not reproduced faithfully, so check any figure you want to use against the tool it came from. It must be symmetric, because A against B has to equal B against A. And it must state its window and its timeframe, or it is a picture rather than a measurement. newyorkcityservers.com names three free tools that publish these grids: Myfxbook's correlation tool, Investing.com's correlation calculator and BabyPips' currency correlation calculator.

How do you work out the correlation yourself?

Two price series over a period you chose, run through the Pearson correlation coefficient.

IG names the method and the route: Pearson "is the most used measure of currency correlations in the forex market, but others include the intraclass correlation and the rank correlation", and "Many traders will use a spreadsheet computer program to calculate the Pearson correlation coefficient, because the method for doing so manually is very complex." logikfx.com shows it done, offering an Excel template where step 1 is to "Copy and paste price data into Data set 1 & 2".

We publish no coefficient of our own. Every figure here belongs to a named source.

Does a correlation hold all day?

No, and IG measured it hour by hour to show it. Its second table runs EUR/USD against USD/CHF from 8am to 9pm UK time, a window chosen to "include the open and close of both the London and New York trading sessions". It starts at -0.967, reaches -0.970 in the late morning and early afternoon, then weakens through the evening to -0.933.

IG gives the reason: "The strength of a currency correlation depends on the time of day, and the current trading volumes in the markets for both currency pairs", with dollar pairs more active during US hours of 12pm to 9pm UK time and euro and sterling pairs between 8am and 4pm. newyorkcityservers.com reports the same effect, that the EUR/USD and GBP/USD correlation "tends to be strongest during the London-New York overlap and weakest during the Asian session." So a daily figure says little about what two symbols did in the two hours you actually trade.

What is the correlation between gold and the dollar?

Strongly negative as a rule, with a known failure mode. Gold is priced in US dollars worldwide, so a weaker dollar makes gold cheaper for buyers holding other currencies, which lifts demand and the price. Dukascopy publishes the exception on the same page: the relationship "isn't perfect and can break down during certain market conditions, such as when both gold and the USD act as safe-haven assets during global market turmoil, causing both to rise simultaneously despite their typically inverse relationship." That failure shows up in the conditions where people reach for gold. IG adds the yen, which "is also widely believed to be a safe-haven currency, and gold is known as a safe-haven asset", so the two tend to move together during uncertainty.

Which commodities move with which currencies?

The exporter's currency moves with the commodity it exports. Dukascopy's table:

CommodityCurrency pairDirection, in its words
Brent Crude OilUSD/NOKNegative. "When oil prices rise, the Norwegian krone strengthens, pushing USD/NOK lower"
WTI Crude OilUSD/CADNegative. "Higher oil prices often lead to a stronger Canadian dollar, lowering USD/CAD"
GoldAUD/USDPositive. "Australia is a large gold producer, so higher gold prices tend to strengthen the AUD"
GoldUSD/CHFNegative. "Safe-haven demand for gold and the Swiss franc can move inversely to the U.S. dollar"
CopperAUD/USDPositive. "Australia's large copper exports result in a positive correlation between copper prices and the AUD"
SilverUSD/MXNPositive. "Mexico is a top silver producer, and higher silver prices often strengthen the Mexican peso"
Iron OreAUD/USDPositive. "Australia's economy benefits from iron ore exports, so higher prices strengthen the AUD"

IG covers three of those in prose and calls the Australian dollar a "commodity currency because its value is tied closely to the value of Australia's commodity exports such as copper, coal, agricultural products and gold". newyorkcityservers.com puts AUD/USD against gold at "+0.60 to +0.80", no period stated.

Am I diversified or just in one trade twice?

If the pairs are strongly correlated, you are in one trade more than once. Three long positions on EUR/USD, GBP/USD and AUD/USD are one short-dollar position in three wrappers. Dukascopy writes that "trading several highly correlated pairs in the same direction efficiently increases risk exposure", newyorkcityservers.com calls it doubling down "on the same directional bet without realizing it", and Benzinga frames the fix as looking for a pair with low or negative correlation to what you hold.

IG frames the same arrangement the opposite way, saying traders "will typically take positions on correlated pairs in order to diversify themselves while maintaining the same overall direction", to protect against "the risk of a single pair moving against them, as they will still have the opportunity to profit on the other pair if that happens". So one source calls three correlated longs a concentration of risk and another calls it diversification, and which it is for your positions depends on a figure for your own window.

How do you check a correlation for the period you are testing?

Open both symbols on the same timeframe, step the same dates on each, and record what the two charts did at every decision point. A table measured over an unstated window cannot tell you whether the relationship held during the weeks you are testing.

It is slower than reading a number off a grid and it answers a different question. A coefficient compresses a window into one figure, so a relationship that held firmly for two weeks and broke in the third comes out looking moderate. Stepping the dates shows which weeks it held.

Charts in ChartLabs open as separate windows, so two symbols can sit on two monitors and the layout can be saved and reloaded. Each symbol reopens on the chart period you last used, so the second one does not quietly reset to a different timeframe. Step both charts yourself and record four things: the two symbols, the period, the timeframe, and whether the relationship held across the whole test or only part of it. That last one is what a coefficient hides, and "held for the first two weeks, broke during the central bank week, came back" is worth more than one figure for the month. The Notes column in the Orders panel and the note button on the chart both take free text, notes come through in the CSV export, and tagging the trades on each symbol lets the analytics be filtered by tag.

The analytics have a Correlations section, and it compares the symbols you traded, combining every selected strategy's trades on each symbol, so it answers whether your own results on two symbols moved together. It is not a price-correlation matrix and will not tell you whether EUR/USD and GBP/USD were correlated over the dates you tested.

Our wider process is in the beginner's guide to backtesting, the session and spread habits that change forex results are in the forex backtesting guide, and why the data has to be clean first is in backtesting accuracy, slippage and data quality.

Check it on your own window

Our historical data comes from Dukascopy, at tick level where tick data exists, cleaned to remove anomalies and built into 1-minute candles, with depth up to around 22 years on the oldest forex pairs and other instruments starting later. One plan, $20 a month or $200 a year, with a 7-day free trial and no credit card required: start the free trial.

Related Topics

forex-correlationcorrelated-pairscorrelation-coefficientforexcorrelation-matrixbacktesting