Pair Trading Strategy Journal

Quote from RogerWilco:

ar1zona, do you need to identify pairs which converge/diverge within one day? If you're talking about regular Z distances (2.0 to track converging, <1.0 to track diverging) - there won't be much pairs like that for sure.

But if you intend to maximize your profits by checking divergence intraday - then you could put an Excel spreadsheet together which will be doing that.

However, I would advice to use it more for paper testing rather then for backtesting, otherwise it will take ages (i.e. approximately 400 times slower on a 1 min data) to back test a single pair.
More...

hi roger, thank you for your reply

im not trying to max profits by taking/exit positions intraday, im sure the EOD results such as PTF are sufficient for swing trade pairs, and this is what i'd like to do once accumulated some capital first

for now what im trying to do is enter and exit positions within the same trading day (preferably not holding the same position from open to close). I agree its quite different for a 1.0 or 2.0 std dev to diverge and converge within a single day, so i want to backtest for a much much smaller divergence to play with. The question is how much Z to trigger a signal, for each pair.

I also agree about how backtesting those tick data will take ages. (I had actually done backtesting ticks in excel, for another strategy, but man that took too long for just a few stocks). Backtesting on minute data appears to be the only choice left, but im not sure how accurate the results will be, since intraday fluctuations within a minute kinda means a lot.. for intraday traders

I know Don Bright's traders do a lot of intraday pairs and they have a system of their own. But for an individual like me, its probably too much work to build an entire intraday backtesting sheet, it would be great if someone somewhere provide this service for a reasonable fee...
 
Since the Standard Deviation depends upon the time frame, I don't understand why one would assume there wouldn't be any 2Z moves on an intraday timeframe.

If we assume that the Standard Deviation grows as the square root of the number of time steps, then there are 86400 seconds in a 24-hour day and the Standard Deviation should be 1/294 as large. If a significant move using a 10-day window is $20.00, then a significant move using a 10-second window is $0.07 .

The bigger issues are finding pairs that have such tight relative value pricing that one can realistically hope to complete the whole trade within a few hours and exactly what intraday timeframe should be used for a specific trade.

What timeframe are intraday spread traders using? 10-second? 10-minute?
 
thanks for everyone's comment

i got the argument about how z-score should be relative to the timeframe now, my bad

and roger, im a prop trader so commission and leverage is not a problem. I guess the biggest cost for trading closely correlated ETF pairs is probably paying the spread, twice, since you have to get in with a market order each time

although i still think there's a way to do it, since a lot of bright's traders do intraday pairs exclusively (read that somewhere), and they have their own intraday backtesting system. Most likely the intraday pair traders will focus on stock though, since ETFs do not offer enough divigence after accounting for spread and ecn fees
 
May be I will share not the brightest idea, but why not to paper test ETF pairs (or stocks) intraday which correlate on a longer time frame ?

The idea is to identify correlating profitable pairs on a 365-day range with let's say 15 days of SD loopback and 30 days of R loopback (for example, with PTF). Then you use intraday tick data to run those pairs through several months but using intraday z-score.

This should be a good enough back test.
 
Quote from ar1zona:

ya somehow the post suddenly dead for no reason... even though everyone's still pair trading

thanks for your reply, btw, do you know any other software/website/service that test intraday pairs?
More...

Not really, you can try pairlog.com but you cant use your own parameters... I now use ptf but also use my buddys program, well I actually ask him to test for me..lol Also some xcel stuff that he has...

But honestly, in the end I still filter all my trades by eye... I stay away from trending ratios usually, I just like to have profitable pairs to start with...
 
Quote from ar1zona:

thanks for everyone's comment

i got the argument about how z-score should be relative to the timeframe now, my bad

and roger, im a prop trader so commission and leverage is not a problem. I guess the biggest cost for trading closely correlated ETF pairs is probably paying the spread, twice, since you have to get in with a market order each time

although i still think there's a way to do it, since a lot of bright's traders do intraday pairs exclusively (read that somewhere), and they have their own intraday backtesting system. Most likely the intraday pair traders will focus on stock though, since ETFs do not offer enough divigence after accounting for spread and ecn fees
More...

From what I've heard many of brights top guys hold overnight pairs..I think for many days...
 
Quote from RogerWilco:

May be I will share not the brightest idea, but why not to paper test ETF pairs (or stocks) intraday which correlate on a longer time frame ?

The idea is to identify correlating profitable pairs on a 365-day range with let's say 15 days of SD loopback and 30 days of R loopback (for example, with PTF). Then you use intraday tick data to run those pairs through several months but using intraday z-score.

This should be a good enough back test.
More...

ya this might work, i found a whole bunch of closely correlated ETFs on marketopology, which someone posted a few pages earlier, and going to get tick data from esignal to test it out

the profit margins for those "over correlated" ETFs are probably too slim tho, and less correlated ones may not converge within an intraday frame, need to find somewhere in the middle
 
You must log in or register to reply here.
Back
Top