Let's discuss academic research on mean-reverting trading strategies...
- Thread starter mizhael
- Start date
Shall need to read more to understand it better.ok this pair trade has stabilized at about -$400, and the passage of time has not cured this. it's a paper trade, i'm closing it for a loss of $400.
possible explanations:
1) i have NOT looked at the ratio of the pair (it's stupid, i know, i know, don't have the software yet). so i did not know where the ratio was relative to its historical levels, whether the pair is trending etc etc
this was only finviz-based. i guess this means if you only look at one stock (instead of the pair in its entirety), you better trade that one stock.
a standalone short Novartis trade would've made some money, but the switzerland etf lost much more, maybe because of problems in the europe etc
2) various etf pairs strategies may work better in the commodities complex, e.g. USO/XLE, GDX/EWC, ABX/EWC etc.
in any event, i am deeply disturbed by this and i think i need to bring myself back into a centered state of mind. i am going to do some breathing exercises and tai chi.
I suggest you download some of the Matlab code to test for cointegration in your pairs. R also has some code and I think Mathematica will do it as well.
What you want is to detect a long term relationship between the pairs. Correlation measures the relationship at a point in time; it is typically an unstable risk factor in financial markets.
Carol Alexander in "Market Models" has a good explanation of correlation and cointegration. She also has some articles on the web that are worthwhile. To find them, just type "Carol Alexander cointegraton" in google.
What you want is to detect a long term relationship between the pairs. Correlation measures the relationship at a point in time; it is typically an unstable risk factor in financial markets.
Carol Alexander in "Market Models" has a good explanation of correlation and cointegration. She also has some articles on the web that are worthwhile. To find them, just type "Carol Alexander cointegraton" in google.
http://www.verbeia.com/mathematica/mathecon/econ_code.htmlhas an implementation of cointegration using johansen procedure.
The author warns it is untested and is freeware. I would take his/her word for it.
The author warns it is untested and is freeware. I would take his/her word for it.
I keep seeing intradaybill mention that if co-integration exists then so does correllation. Nothing could be further from the truth.
People who pairs trade trade co-integration! They do not trade correllation!
In general if you find a pair that is co-integrated, then most likely it will be correlated but this is not always true and in fact you do not want it to be true.
The best pair to trade is one that is co-integrated (i.e. stationary) but is negatively correllated.
Great description. Too bad some software vendors won't read your comment.
A very good quote, Ms Varima-Garch. It spoke a lot of how greed blinds our thinking.
The correct link is as follows:
http://www.verbeia.com/mathematica/mathecon/econ_code.html
That is completely WRONG!
Read the last section here:
http://www.financialwebring.org/gummystuff/correlation-vs-days.htm
and
http://epchan.com/downloads/cointegration.pdf
enjoy...
[Edit] See marine's excellent post above...
It is true that correlation does not imply cointegration. I think we can agree to that.
Cointegration implies correlation when the residuals are stationary. When the residuals are normal, there is longer term correlation.
In the paper you posted there is a basic error in that if the residuals are random the two series are not cointegrated and you cannot know that from a limited sample.
I thus insist, cointegration implies correlation, but not vice versa.