Cointegration of currency pairs | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Currency Pairs - Same Currency, Different Direction? 7 replies

Synthetic hedges, cointegration, mean reversion and similar stuff 419 replies

Trading the cointegration 31 replies

Pairs of Currency Pairs 4 replies

Cointegration, Synthetic hedges, mean reversion in R, Tech Thread 0 replies

Attachments: Cointegration of currency pairs
Exit Attachments

Cointegration of currency pairs

  • #1
  • First Post: Mar 26, 2008 3:18pm Mar 26, 2008 3:18pm
How well are currency pairs Cointegrated as opposed to correlated? Has anyone done any work on this subject or could pass on some links?

Based on my observation, it would seem that correlated pairs (EURUSD vs USDCHF for example) tend to trend in the direction of the interest rate differential over an intermediate term (daily/weekly) as opposed to mean revert. Any comments, etc.?
  • #2
  • Mar 26, 2008 3:54pm Mar 26, 2008 3:54pm
Have a look at this you should be able to find what you are looking for

http://www.forexhit.com/analyze-fore...relations.html
  • #3
  • Mar 26, 2008 4:29pm Mar 26, 2008 4:29pm
Quoting Kiwi06
Have a look at this you should be able to find what you are looking for

http://www.forexhit.com/analyze-fore...relations.html
Hi Kiwi, thanks for the link.

Unfortunately this shows correlation which is different from COINTEGRATION.

http://www.themoneyblogs.com/quanttr...rrelation.html

Check this out for further information...
  • #4
  • Mar 27, 2008 10:07am Mar 27, 2008 10:07am
are you talking about negative correlation?

so when one goes down the other goes up and vice versa?
  • #5
  • Mar 27, 2008 10:31am Mar 27, 2008 10:31am
No, I'm looking for cointegration. I would like to apply some typical stat arb approaches to FOREX to see if there is any excess return.

Think of Cointegration as a measure of mean-reversion. The more two time-series is cointegrated, the greater the tendency for the two series to revert back to some mean, either 0 or some other average/trend/value etc.

Its possible or two securities to be correlated but not necessarily cointegrated. In the stock world, most equities are correlated to some degree - as market goes up so do the stocks but the spread doesn't necessarily follow a pattern.

Just seeing if someone had done the work, no point reinventing the wheel. I put on a pair yesterday EURUSD vs USDCHF in a demo, picked up 160 pips this morning. This pair seems as if it mean reverts to a basic trend, I'll probably have to utilize a regression line and look at the deviations from that....

Anyone actively pairs trading currency pairs...??
  • #6
  • Mar 27, 2008 10:44am Mar 27, 2008 10:44am
i think USD/CHF and EUR/USD are negatively correlated

just put them in a clear simple sense

when one goes up the other goes down
Attached Image (click to enlarge)
Click to Enlarge Name: negative correlation.gif Size: 57 KB
  • #7
  • Mar 28, 2008 8:01am Mar 28, 2008 8:01am
Quoting Gift_Art
i think USD/CHF and EUR/USD are negatively correlated

just put them in a clear simple sense

when one goes up the other goes down
Hey Gift Art,

Please read this link, it can probably explain better then I can:

http://www.themoneyblogs.com/quanttr...rrelation.html
  • #8
  • Mar 28, 2008 9:01am Mar 28, 2008 9:01am
Quote
Think of Cointegration as a measure of mean-reversion. The more two time-series is cointegrated, the greater the tendency for the two series to revert back to some mean, either 0 or some other average/trend/value etc.
Well sort of. Prices won't wander a fixed "maximum distance from a mean" really, and it's not like you can just say "ok, they're 200 pips from the mean, there's a huge probability it's going to regress now!" because you're not factoring in enough details. But you can use averages (and the distance to them) to determine the strength (and direction to an extent) of a move.

Long story short it all boils down to support and resistance. Price needs a rising support to keep rising, or a falling resistance to keep falling. Pairs won't regress to the exact same mean, but if you start plotting moving averages (21 EMA/SMA, 25 EMA/SMA, 50 EMA/SMA, 62 EMA/SMA and a number of others) and start looking at what correlated pairs do when they hit those averages (or get really far away from them) you'll see a pattern. It's not magic or even math, it's just reinforced behavioral psychology.

But I don't think you can predict that "max distance" with any degree of accuracy even by comparing relationships between pairs or the historical/statistical deviation from past distances. Every moment in the market is unique and all that jazz. At best all you'd come up with is yet another "occasionally accurate" indicator.

Still as basic elements go this is a very common trading strategy. Some people play the regress (I did last week on the GBPUSD), but most play the bounce off the average since it's easier to catch. Look at the James16 chart thread and see how he uses the moving averages there.

What you'll find is that just using MA bounces or regressions by themselves is seriously prone to whipsaw, especially during ranging markets, but with some discretionary price action and fib work you can play a "weight of the evidence" method and do quite well. I anticipated most of the ranging conditions this week (cable, euro and yen) by comparing the weekly and the distance from averages... so it does work.
  • #9
  • Mar 28, 2008 10:47am Mar 28, 2008 10:47am
i understand what you are trying to say it is just EASIER to think that it is just a negative correlation, nothing more

let say use the 200 SMA for e.g.

if price moves several pips away from 200 SMA eventually they will revert back to the MA and now we need to find 2 pairs which have exactly the opposite momentum

so if pair 1 goes 200 up ABOVE the 200 SMA
pair 2 goes 200 pips BELOW 200 SMA

and at some points of time they will return to 200 SMA = the mean state = the equilibrium = the waterlevel

the way to statistically calculate this probability is by doing the correlation math, and we will have a correlation of -80 (for eg) and that's what we have with USD/CHF and EUR/USD

now if you take a closer look of my attached graph you will find your EDGE quite soon, just make everything simple in forex
  • #10
  • Mar 28, 2008 10:57am Mar 28, 2008 10:57am
Quoting Rabid
Well sort of. Prices won't wander a fixed "maximum distance from a mean" really, and it's not like you can just say "ok, they're 200 pips from the mean, there's a huge probability it's going to regress now!" because you're not factoring in enough details. But you can use averages (and the distance to them) to determine the strength (and direction to an extent) of a move.

Long story short it all boils down to support and resistance. Price needs a rising support to keep rising, or a falling resistance to keep falling. Pairs won't regress to the exact same mean, but if you start plotting moving averages (21 EMA/SMA, 25 EMA/SMA, 50 EMA/SMA, 62 EMA/SMA and a number of others) and start looking at what correlated pairs do when they hit those averages (or get really far away from them) you'll see a pattern. It's not magic or even math, it's just reinforced behavioral psychology.

But I don't think you can predict that "max distance" with any degree of accuracy even by comparing relationships between pairs or the historical/statistical deviation from past distances. Every moment in the market is unique and all that jazz. At best all you'd come up with is yet another "occasionally accurate" indicator.

Still as basic elements go this is a very common trading strategy. Some people play the regress (I did last week on the GBPUSD), but most play the bounce off the average since it's easier to catch. Look at the James16 chart thread and see how he uses the moving averages there.

What you'll find is that just using MA bounces or regressions by themselves is seriously prone to whipsaw, especially during ranging markets, but with some discretionary price action and fib work you can play a "weight of the evidence" method and do quite well. I anticipated most of the ranging conditions this week (cable, euro and yen) by comparing the weekly and the distance from averages... so it does work.
here's my answer to you...

YES we can "measure" the max distance I have proved this in forex and stocks and they are measureable to some degree. of course this is not the holy grail or 100% bs. you are on the right thinking track though.

this is a very important concept, knocks420 is on his way to become a good trader ... quite soon i believe
  • #11
  • Mar 28, 2008 4:16pm Mar 28, 2008 4:16pm
Quoting Gift_Art
here's my answer to you...

YES we can "measure" the max distance I have proved this in forex and stocks and they are measureable to some degree. of course this is not the holy grail or 100% bs. you are on the right thinking track though.

this is a very important concept, knocks420 is on his way to become a good trader ... quite soon i believe

haha, thank you for your kind words Gift Art! I hope I'm already there, I work for a MM+ hedge fund, I sure hope I know how to trade!

Our stat arb strategies are all focused on equities and there is most definetly a statistically significant measure of cointegration between equity pairs. I.E. there is a level beyond a mean where prices are highly likely to revert. Now that mean is not necessarily a spread to the SMA, its usual a deviation based upon past historical movement and our own manipulation of the time series. Can you generate further alpha based on fundamental/technical analysis of the underyling, you bet! Can you convince the PM, probably not! lol.

Unfortunately equity stat arb is overflowing with players squeezing returns. FOREX is attractive due to the liquidity, trading hours, spreads, commish, etc. I would assume its likely pretty saturated as well but our smaller size is a benefit in this situation.

Just curious to see if someone had to done the hardwork for me, haha!
  • #12
  • Mar 28, 2008 6:56pm Mar 28, 2008 6:56pm
Quote
YES we can "measure" the max distance I have proved this in forex and stocks and they are measureable to some degree. of course this is not the holy grail or 100% bs. you are on the right thinking track though.

Really... hmm, I've never seen any way to do that. You're going to have to back that up, lol. How can you predict the maximum distance from an average with any edge?

Take a look, for instance, at the JPY pairs. The USDJPY was in a free fall and reached a pretty large distance from the 21 MA. It dropped right till the 100, then retraced back. Easy to predict from a fundamental perspective, how can you predict that from a deviation from the mean perspective?

Now compare that with other JPY pairs. The other pairs are all much tighter with the average due to stronger fundes. How can I use cointegration from those pairs to determine an "overbought" or "oversold" condition?

Quote
Our stat arb strategies are all focused on equities and there is most definetly a statistically significant measure of cointegration between equity pairs. I.E. there is a level beyond a mean where prices are highly likely to revert. Now that mean is not necessarily a spread to the SMA, its usual a deviation based upon past historical movement and our own manipulation of the time series. Can you generate further alpha based on fundamental/technical analysis of the underyling, you bet! Can you convince the PM, probably not! lol.

So not a moving average, but another average? There are an infinite number of ways to compute an average. So you're saying there's a way to compute an average and determine at which distance from that average price historically regresses and that such a number provides a statistical edge (yielding a greater than 50% success rate on entry calls) on future regressions?

I could easily back test and chart distances from an average over time and get numbers... not so sure they'd be any long term use. Well, no more than a traditional stochastic indicator. But if there's another average or approach I'm missing... that'd be nice to learn.
  • #13
  • Mar 28, 2008 7:15pm Mar 28, 2008 7:15pm
Quoting knocks420
haha, thank you for your kind words Gift Art! I hope I'm already there, I work for a MM+ hedge fund, I sure hope I know how to trade!

Our stat arb strategies are all focused on equities and there is most definetly a statistically significant measure of cointegration between equity pairs. I.E. there is a level beyond a mean where prices are highly likely to revert. Now that mean is not necessarily a spread to the SMA, its usual a deviation based upon past historical movement and our own manipulation of the time series. Can you generate further alpha based on fundamental/technical analysis of the underyling, you bet! Can you convince the PM, probably not! lol.

Unfortunately equity stat arb is overflowing with players squeezing returns. FOREX is attractive due to the liquidity, trading hours, spreads, commish, etc. I would assume its likely pretty saturated as well but our smaller size is a benefit in this situation.

Just curious to see if someone had to done the hardwork for me, haha!
errrrr not that complicated though you don't need anything more than MA

again i repeat

it's all in the picture I posted if you can't see what i see print it and let someone else look for you perhaps their simple mind can see the EDGE through

rabid, now you get the wrong thinking process, read your answer to me once more there is a big flaw, just look at the picture i posted
  • #14
  • Mar 28, 2008 7:20pm Mar 28, 2008 7:20pm
btw i'm using MACD 12 26 9

but my theory works with other indicators as well on both forex and stock charts

i have tested the following
S/R, EMA cross they are all working

so the key in trading is not to find the best indicators but to interpret them correctly
  • #15
  • Mar 28, 2008 7:52pm Mar 28, 2008 7:52pm
Quoting knocks420
haha, thank you for your kind words Gift Art! I hope I'm already there, I work for a MM+ hedge fund, I sure hope I know how to trade!

Our stat arb strategies are all focused on equities and there is most definetly a statistically significant measure of cointegration between equity pairs. I.E. there is a level beyond a mean where prices are highly likely to revert. Now that mean is not necessarily a spread to the SMA, its usual a deviation based upon past historical movement and our own manipulation of the time series. Can you generate further alpha based on fundamental/technical analysis of the underyling, you bet! Can you convince the PM, probably not! lol.

Unfortunately equity stat arb is overflowing with players squeezing returns. FOREX is attractive due to the liquidity, trading hours, spreads, commish, etc. I would assume its likely pretty saturated as well but our smaller size is a benefit in this situation.

Just curious to see if someone had to done the hardwork for me, haha!
knocks i really hope you dun think i'm in anyway underestimated your trading skills, but i believe that to trade is to continuously open our eyes and learn. the market will shift and i know that and whether the fed rate is 0.5% or 20% i dun really care, i will just go where the market wants me to go

i saw you in brian's thread and your question there was actually what led me to this thread i didn't even know you started this one

  • #16
  • Apr 1, 2008 8:18am Apr 1, 2008 8:18am
Gift Art,

No offense taken, it's just an anonymous message board, no reason to take offense at anything on one, lol. Take a look at a longer term spread between EURUSD, USDCHF. Trading a mean reversion would've been right eventually but man o man that would've been a lot of pain. There were likely opportunities in a shorter time frame however you can see that it was probably more profitable and easier to trade with the spread as opposed to against it!

I think one of the biggest problems we all have, and especially small retail players, is extrapolating one or two occurences into infinity. Professionals do it as well, thats why its important to have a large sample of data and look for occurences where your hypothesis DOES NOT happen. Then analyze they why's, what ifs, risk management, etc.
Attached Image (click to enlarge)
Click to Enlarge Name: EURUSD USDCHF.JPG Size: 319 KB
  • #17
  • Sep 25, 2008 9:40am Sep 25, 2008 9:40am
I think I would be more interested to see if there was cointegration of government backed securities of countries and their currency pairs...Seems like they would 'wander' seems like they would wander back and forth a lot more opening up more trading opportunities.

Say you have country A, B. Are (bond A)/(bond B) cointegrated with currency pair A/B? I plan to look into it...any thoughts anyone?
Kiss'n all the way to the top!
  • #18
  • Dec 30, 2008 4:39am Dec 30, 2008 4:39am
I read this thread and I can't stop laughing how hard is to understand the different from cointegration, correlation and the spread from of a sma...I'll try a bit later to explain what cointegration really is
"Abandon all hope, you who enter here" La Divina Commedia, Dante Alighieri
  • #19
  • Edited 6:03am Dec 30, 2008 5:50am | Edited 6:03am
According to Wikipedia: Cointegration is an econometric property of time series variables. If two or more series are themselves non-stationary, but a linear combination of them is stationary, then the series are said to be cointegrated.

Here is a little chart to show the concept, using the euro and the pound's time series between March and September 2008. And yes, you can trade based on this model I build in 2 mins...

Attached Image (click to enlarge)
Click to Enlarge Name: cointegration.jpg Size: 36 KB
"Abandon all hope, you who enter here" La Divina Commedia, Dante Alighieri
  • #20
  • Dec 30, 2008 7:27am Dec 30, 2008 7:27am
Very interesting indeed. Can you explain a bit more the "model you have built" ? Not that I feel ready to go for it, as i still have some work to do to understand why cointegration trading opportunities are that interesting but I definitely want to get a good understanding of this "tool".
4 traders viewing now