Cointegration of currency pairs

that model is nothing more than the regression residuals (the difference between the real value and the foretasted one).

Just google cointegration, or similar terms

ps. here is a nice article explaining the difference between cointegration and correlation

http://www.tradingmarkets.com/.site/...orrelation.cfm

hi semar

can you give me an example with metatrder chart how you would measure if two pairs are cointegrated or not

if its easy lilke this than i look forward

thanks

Great information!
I’ll join lodol2 in that request.
I would be nice an example whit a MT4 chart.
Regards

Lodol,

I use a routine in MatLab that runs cointegration across a basket of securities. I actually like the previos posters idea of cointegrating FX with Money Market rates or Govy Bonds.

The alternative to cointegration is to just perform a test of a strategy on a given pair. Since there are only a limited number of pairs, this shouldn't be too difficult and apply some common sense to the outcome.

Unfortunately I do not use MT so I can;t help with the coding.

ok thanks knocks

so it seems that i have to use mathlab for this....

can you tell more about your mean reversion trading style

thanks

lodol

theres not much to say, I look to fade extreme moves on intraday time frame. Focus on EURUSD due to spreads. Works great but I'm really looking for solid trend-following models to compliment the MR trading Edit: I should say I've tested a lot of the ideas in the trading discussion thread and almost none work consistently enough to implement mechanically.

hello

would be great if you could show with an chart some of your trades.(if times allow)

do you use indicators or only prive action to determine the reversal and how you define an inbalance which will probobly return the price to mean..?

thanks

regards

Cherry-picked good day:



Cherry-picked bad day:



As you can see this is why I am interested in a trend-following strats.

hi knocks

of course in a trend you want to have trendfollwoing and inranging you are better with reversal strategy....of course smile....

but for me the logic of "price will revert to its mean" made all t he time the most sense....

i thought to find trendfollwoing systems is a easy task ..........because most are like this, but of course you need also a trend then..........................................laughing

but you said in one post that you prove most of the systems and not find one which reach your criteria........

is it not to find out "in what state the market is" to trade "trending or reversal" ?

so the point would be is it possible to filter out in which state is the market is.... then also quite simple systems would work? or?

regards

lodol, Yes, if I have understood you correctly,I think the 'grail' is to be able to identify trending/reverting regimes before they happen and trade accordingly. The trick is how do you forecast this without giving up too much profit. Quantitatively this hasn't been done (it may have been accomplished by individuals who have large financial incentive not too share though [Renaissance maybe?]) I would venture that discretionary traders seem to have a handle of this better then most. There have been some papers regarding volatility regime changes but I haven't done much research myself. My latest project is to try and replicate a multiple MA system on a portfolio of pairs which is being discussed in another thread. It is taking me awhile to port the data so results are still forthcoming...I will say on my first test, it is profitable to fade traditional MA crossover systems.

(off the subject of cointegration)

I'm wondering (out loud), it has been said by market writers, that markets range about 80% and trend 20% of the time. A mean reversion algorithm and position management strategy could be combined to exploit the theoretical "20%" trending periods. Some published articles have advocated fading the market with multiple entries, (attempting to mimic dealer behavior), using this concept, one could hold onto the most advantageous entry, entered using mean reversion, and protected with a stop, to exploit a "newborn" trend.

One could then devote time to developing the best suited reversion strategy for fx, instead of attempting to address both market conditions.

Using the definitions of the concept from here and here, the attached indicator for MT4 attempts to display cointegration data from the pairs under consideration similar to this.

Very interesting. I currently use a toolbox in MatLab to perform cointegration analysis.

I am not as familiar with coding in MT4; if you have time could you possibly code a script to perform a simple pair trade with various securities and time frames? PM me if you need assistance with the logic on a pairs trade.

I looked at this code and this is merely measuring the spread, it is not a test of co-integration. Granted if a particular spread results in a profitable pair then it is likely co-integrated.

You can't have read the links very well, testing for co-integration even in its simplest form would involve running a regression of one price against another then testing the residuals for mean reverting behavior, not simply subtracting the prices.

sign of intelligence on this forum???



lol, jk

Hi knocks420 what matlab toolbox do u use for the coint tests?

The financial toolbox includes Johansen Procedure for co-integration I believe. It has been awhile since I actually ran the script so I could be mistaken.

Seems that not many people even know what cointegration means. I didn't either about a year back, since then i have developed co-integration analysis in Matlab for forex or NYSE. I have gone as far as writing a script to run cointegration on the whole 9,000,000 combination of stocks on NYSE, in about 8hrs. Similar to a hedge fund? My reseach shows, hedge funds do global analysis in about 4hrs. Forex is easier since there are limited number of time series. If you are serious about developing co-integration analysis software, buy a student version of Matlab, and download this free toolbox specific for cointegration http://www.spatial-econometrics.com/. They have a complete set of econometric analysis functions, and at first it is going to be confusing, but study ADF and CADF functions, that should give you a huge start i didn't have. It took me a while to understand how to interpret the results. My mistake is: doing analysis on high frequency data, and never actually trading the research of cointegration pairs. I might go back to it a lil later, but co-integration does exist between currency pairs, and it is not necessary USDCHF and EURUSD, you will be surprised which pairs have mean-reversion. If you are interested i can sell you the code that i have developed, it does everything for analysis for entry & exit conditions, as this strategy is exactly the same as a hedge correlation strategy, except the analysis is a bit different.

Nice post, this is all great stuff but the problem as I see it is that you are going to fall into the same data mining bias traps. If you run tests for co-integration over any large set of data you are bound to find series that exhibit mean reversion, how do you differentiate the pairs which are co-integrated for a specific reason from those which are simply statistical artifacts generated by the search process itself? Also, standard co-integration packages only test for static co-integration where the hedge ratios are constant, this is a fair limiting concept if one thinks about it.