Pair Trading Strategy Journal

Quote from Appleseed:

I may be entiely wrong in this asumption but i'll go ahead.

I have a feeling that the winning % is just curve fitting. You can
find hundreds of pairs that are in the winning 80 to 90 percentile
today and going foreward will not maitain that level of a winning percentage of the past. If you select a group of winning pairs today and go forward 6 months with the same consistancy of 80 -90% winners you may have good candidates for pair trading.
Any thoughts??

cheers
john
More...

John,

I should probably think more about your post before answering, but your post raises questions that I had been pondering also.

For example, just because I filter my universe of pairs based on correlation and %/$ profit, does that mean I am cherry picking the best pairs going forward, or just the coincidental, chance favorites over the last 6 months. I agree with you that perhaps these may not be in the top of the list in the next six months, based on the same filtering technique.

Perhaps I might be better taking those correlated pairs (I sort of believe in correlation as a method of identifying equities that have historically followed the same path) that have a poor profit history - the ones below the mean $ return, thinking that these pairs may (using the same strategy as pairs trading itself) show better returns as their $$ returns move back to the mean.

I'm not sure if I agree with the above statement, however cherry picking the historically best winners might be flawed.

But then again, the Turtle traders made millions trend trading. Picking the winners is following a trend, one that may not yet be fully exploited, and thus still have a good edge.

Which is right? Neither until proven correct, and the only way is to test and see. Jonny picks those that show a $ profit above $300 (unless he has since changed his filters), and has shown the trend style works. Perhaps picking the low $ winners historically works better? I might try sim trading this method to see if there is an edge here.

I'm probably too verbose for a saturday afternoon. Sorry... :-)

Have a good weekend all.

Adrian
 
Good points, keen to see peoples

Also, once I build my universe of pairs, the database remains quite static. There may be many great quality pairs that showed low correlation / loss win $ ect ect during my backtest and I have missed. While others that showed great stats during my backtest infact have broken down in correlation or mean reversion.

What would work perfectly is if PTF had some sort of system that continually backtested categories and pairs.......then automatically uploaded ceratain pairs into the console (or set off some sort of alert) as they came began posting strong stats.... Existing weaker pairs would drop off in order to make room......Sort of like an PTF S&P500 or something


Quote from cipherscribe:

John,

I should probably think more about your post before answering, but your post raises questions that I had been pondering also.

For example, just because I filter my universe of pairs based on correlation and %/$ profit, does that mean I am cherry picking the best pairs going forward, or just the coincidental, chance favorites over the last 6 months. I agree with you that perhaps these may not be in the top of the list in the next six months, based on the same filtering technique.

Perhaps I might be better taking those correlated pairs (I sort of believe in correlation as a method of identifying equities that have historically followed the same path) that have a poor profit history - the ones below the mean $ return, thinking that these pairs may (using the same strategy as pairs trading itself) show better returns as their $$ returns move back to the mean.

I'm not sure if I agree with the above statement, however cherry picking the historically best winners might be flawed.

But then again, the Turtle traders made millions trend trading. Picking the winners is following a trend, one that may not yet be fully exploited, and thus still have a good edge.

Which is right? Neither until proven correct, and the only way is to test and see. Jonny picks those that show a $ profit above $300 (unless he has since changed his filters), and has shown the trend style works. Perhaps picking the low $ winners historically works better? I might try sim trading this method to see if there is an edge here.

I'm probably too verbose for a saturday afternoon. Sorry... :-)

Have a good weekend all.

Adrian
More...
 
Quote from cipherscribe:

Some good information to think about. I guess it really comes down to whatever works for the individual. Certainly cointegration seems to be the valid method from a mathematical perspective, but correlation seems to be a reasonable facsimile, going by the results of Jonny's trades.
More...


I disagree.
Johnny's results just prove he's able - with a ton of good subjective judgement - to extract some good trades from a ton of potential candidates popped up by a
data mining software.

They don't prove correlation is a good proxy of cointegration...

I'm not a quant, but they are differents concepts (*) and this is another of the best explanation I've found on the web, without any disturbing maths ( the latter part is the signifuicant one):
http://financialwebring.org/gummystuff/cointegration.htm

(*) I've already tried to raise this point, but this is a trader's journal and - understandable - concepts draw less attention than the actual BUY&SELL's routine:
http://elitetrader.com/vb/showthread.php?s=&postid=2423148#post2423148
http://elitetrader.com/vb/showthread.php?s=&postid=2423983#post2423983
 
Quote from total_keops:


And I am ont sure the Johansen procedure is the right one for pairs trading!
More...


It surely is, but it is just one of the methods available, and certainly not the easiest starting point for the average person.

Take a look at these (the second link is more undestandable than the first, isn't it?):

http://en.wikipedia.org/wiki/Johansen_test.

http://en.wikipedia.org/wiki/Cointegration

Just out of curiosity, where did you find this spreadsheet?
 
hi guys, i'm trying to learn more about pairs

can anyone comment on this rookie idea pls:

i'm using marketrac.nyse.com/mt/index.html , a visualization tool from nyse, to get a feel for what's going on in the market. it shows DJIA components

it's 20 min behind actual market prices

if i see a stock that's way out of balance, can i take it as a signal, at least an initial signal, to trade it against the dow jones?

(i mean it may look like it's out of balance, but it may actually be reverting to some mean ratio value, so that wouldn't be tradable, i understand that)

can i trade the stock that's out of balance with DJIA against, say, DIA, the dow etf. e.g. XOM against DIA? (i'm trading etfs/stocks only)

if this sounds silly, please let me know. thanks!

Varima
 
I just downloaded the trial version of PTF. I've only looked at the software briefly and had some questions.

What is the basic method they are using to flag new pairs, and to close out positions? Is it basically a z-score on the ratio of the pairs?

Another question is can you do a long term backtest on the pairs? I like to view the cumulative open equity going back many years but the backtests I saw only went back a year or so(?)
 
Another question I had was about indexes.

I added some stock indexes from around the world and saw some decent pairs.

But I'm wondering what price is the platform using to enter trades? If same date close price then that's an issue.

For example if I'm trading HangSeng/DJI pair, HangSeng is already closed but the simulator gets me in anyway.
 
Quote from Angelo_60:
Just out of curiosity, where did you find this spreadsheet?
More...
I was just Googling about cointegration and trying to find a C++ code but I dont remember where.
 
Quote from cooper1308:


I have a question for Johnny,

How much emphasis can I put on winning %? Often i'll find stocks that correlate 40-50-60% however have a win ratio of 97-100% on over 12 trades! with win/loss ratio of 2+

Obviously the more supporting indicators the bigger the edge but how legit is this winning % coupled with win/loss ratio?

If I were to build a universe with say 500 pairs with winning % of 95% and win/loss 2+....would I be correct in allowing for pairs with lower correlation?
More...

You have to be careful with optimizing, its a double edged sword, ideally you want to include pairs that have good historical results but you can't just hand pick the best ones because past performance is not always indicative of future performance, yes I look for pairs that have performed well, I take the view if a pair produces good profits then it comply's with the logic of the system, however i stick to strict filtering criteria to choose pairs not just all the top profit producing one's, and yes high correlation is one of those, there are plenty of low-correlated pairs that make good profits however I believe the inherent risk in those pairs is more than well correlated pairs.
 
You must log in or register to reply here.
Back
Top