I like this idea however I find getting this thing started is incredibly confusing for someone not very familiar with R or mt4. Could someone please post steps to get this project up and running in terms that are simplified for non-regular R and MT4 users? It would be much appreciated. I'm having problems getting charts up on R and I'm pretty certain I've followed the instructions. When I drag the arbomat on the MT4 chart, it just opens R and nothing is there.
System to hedge every major dollar pair for profit / low DD
- #701
- Edited 2:00pm Oct 28, 2012 1:26pm | Edited 2:00pm
- Joined Jun 2009 | Status: A slave to the tape | 4,390 Posts
- #702
- Nov 8, 2012 4:05am Nov 8, 2012 4:05am
- Joined Jan 2011 | Status: Senior Trader | 1,475 Posts
Bulls are stupid Animals!especially when Im short!
- #703
- Nov 8, 2012 5:10am Nov 8, 2012 5:10am
- | Joined Jul 2009 | Status: Trade. Review. Improve | 1,018 Posts
- #704
- Nov 8, 2012 6:45am Nov 8, 2012 6:45am
- Joined Jan 2011 | Status: Senior Trader | 1,475 Posts
Bulls are stupid Animals!especially when Im short!
- #705
- Nov 8, 2012 7:46am Nov 8, 2012 7:46am
- | Joined Jul 2009 | Status: Trade. Review. Improve | 1,018 Posts
- #707
- Dec 6, 2012 7:05am Dec 6, 2012 7:05am
- | Joined Jul 2009 | Status: Trade. Review. Improve | 1,018 Posts
- #709
- Dec 6, 2012 7:21am Dec 6, 2012 7:21am
- | Joined Jul 2009 | Status: Trade. Review. Improve | 1,018 Posts
- #711
- Dec 6, 2012 7:26am Dec 6, 2012 7:26am
- | Joined Jul 2009 | Status: Trade. Review. Improve | 1,018 Posts
- #714
- Mar 20, 2013 7:56am Mar 20, 2013 7:56am
- | Joined Jul 2009 | Status: Trade. Review. Improve | 1,018 Posts
- #718
- Edited 2:52am Mar 22, 2013 2:39am | Edited 2:52am
Sensible points, the OOS issue seems to be universal when it comes to EA development, lot of predictions algorithms fail for the same reason of in sample over-fitting. The PairTrading R package looks fine for me and I am playing it to try if it can work as a indicator (load data of a pairs and show trading simulation results), but indeed I am not sure if it helps in live trading when we talk about real money and future spread dynamics.
FXEZ, can you shed some light in pair trading or statistical arbitrage in general that an average individual...
Actually I consider myself to be an average individual as described. However, let me add right at the outset that I'm not trading based on cointegration. I studied the topic in depth but found that I was in fact a short term guy rather than a long term guy. I don't offer this to influence your decision in any way, (correlation has its own stack of problems) just by way of disclosure that in this context I'm more the teacher who teaches rather than the one who trades the cointegration concept.I'm not sure what I can add to what I've already posted in this thread. Basically speaking if you are interested in the stat arb concept this thread and the related links are a good place to start. The original thread (arbomat) has lots of interesting contributions and I would strongly suggest reading that whole thread. Also take a look at Old Dog's Taming the Beast thread to get a reality check on the cointegration concept from a sound scientific perspective. There is a good back and forth between Old_Dog and 7Bit that properly exemplifies the scientist vs. trader arguments respectively. Choose wisely whose views you heed!
At the outset, you will need to choose a platform and determine if you are a correlation person or a cointegration person. Correlation is more appropriate for short term relationships and cointegration for long term relationships so your specific goals will play into that. (That said, there have been isolated reports by some who are using cointegration from a short term perspective. I never got this to work but that doesn't mean some aren't doing this very thing.) R, Matlab and Python have capabilities for pairs trading. On this site arbomat was written as a proof of concept for R with pairs trading. It approaches the subject from a trader's point of view rather than a scientist's point of view. If you use Metatrader, I suggest taking Arbomat out for a spin to get a feel for how it works then studying the code to determine how to interface to R from Metatrader or the proper R commands to perform a simple cointegration test. Then if you like the PairTrading package (I haven't used it) you can integrate those commands into Arbomat using Arbomat as a template for viewing the pair relationships in Metatrader.
The words of saussiche are worth repeating as the temptation is to attempt to fit anything with no real thought regarding reasons why there should be co-movement:
This is why people should be looking for statistical relationships between instruments where there is some logical reason at work. That's why pair traders tend towards trading equities, because they can find pairs or baskets that share some real world relationship, like gold and mining companies, coffee and coffee shops, or airliners and jet fuel (although the coffee one probably wouldn't work)
For forex I would try a basket of the best/basket of the worst swap pairs (AUDCHF, NZDJPY vs USDZAR, EURAUD etc.), I bet the spread would be pretty easy...
In particular If I were to approach the subject again I would take a hard look at the consistency factor as one of the main ways for judging fit of cointegration (over ADF / Johansen statistical tests) and validity of the model building process. Study the graphics posted by Mechthildche on the Arbomat Thread. Note that principal components is preferred to linear model (lm) for stability as illustrated in those pictures on that link. Also if you choose to go with R and you're new to the platform, you really should take a look at Analysis of Integrated and Cointegrated Time Series With R as a way to walk through the model building process via the urca package with some additional information on the vars package. But there are numerous resources and packages for Matlab and Python with pairs trading so those routes are good paths as well.