Pair Trading Strategy Journal

[tradingtrading]

Quote from CBuster:

sure. i'm not a trading (or pairtraiding) naysayer. i trade full-time, including some pairs stuff. risk is all part of the game. my only question is whether pairs like this particular one (ebay / ostk) represent good risk reward. but that's down to the individual i guess.

When I saw OSTK/EBAY mentioned, I was initially surprised because I know a bit about the history of Overstock. I won't go into it, but you can read here...

http://en.wikipedia.org/wiki/Patrick_M._Byrne

... so it's no surprise to me that OSTK share price was being moved around by extraneous factors other than those it might be sharing with ebay.

Of course, those reasons may well have dissipated and that matbe what the more recent price action is showing you. I suspect so.

Nevertheless, I think CBusters point is a good one. Any form of trading shouldn't just be about the statistical relationships.

 

[jonnysharp]

I don't want to get in a heated debate about co integration vs correlation, as I can see the merits of both, and as some have mentioned co integration is more complicated and a bit harder to understand, however most of us here are using correlation and it seems to be working fine. Can someone here please show an example of a highly correlated pair which has low co integration? would be interesting to look at.

Like someone said you can't rely purely on the statistical relationship to cover your back, you gotta apply some common sense, that's what makes a good trader, identifying situations the computer can't.

 

[total_keops]

Quote from tatankas:


Also, I doubt it's possible to calculate co-integration in mili-seconds, like you calculate correlation.
When you need to analyze hundreds of pairs, the pairs indicators should be calculated on the fly.

Dont worry about that. Everyday I take over 1550 stocks and analyse all possible pairs combination (over a million). I can do that in about 40 seconds with an unoptimised code.

 

[renton]

Quote from total_keops:

Dont worry about that. Everyday I take over 1550 stocks and analyse all possible pairs combination (over a million). I can do that in about 40 seconds with an unoptimised code.

Hi total - how do you do this? I've been looking at PTF 30 day trial for a few days now (no trades but simply curious) and it seems to take a good 10 mins to spot pairs given about 20 Stock Symbols (running on a mid-range PC)

Are you using PTF as well? Or simply your own algorithms / programs?

 

[total_keops]

Quote from renton:

Hi total - how do you do this? I've been looking at PTF 30 day trial for a few days now (no trades but simply curious) and it seems to take a good 10 mins to spot pairs given about 20 Stock Symbols (running on a mid-range PC)

Are you using PTF as well? Or simply your own algorithms / programs?

I have my own code that I wrote in C++ probably much faster than the VB that they might be using in PTF. I can use it with 1 minute bars and scan once per minute for intraday patterns. I guess PTF is fine for daily things.
I do it with the correlation, not the cointegration. Since we are looking at pairs in a trading range I think the correlation is a good estimate of the relationship. Maybe not good enough to make a reliable automated program but since we are applying a human filter to the trades it's fine.
Cointegration is not the latest deal in pairs trading. Modeling dependance through copulas is what's the latest cool gear. And later there will be dynamic copulas, not yet in the research papers but it's coming for thoses who are interested.

 

[Angelo_60]

Quote from jonnysharp:

I don't want to get in a heated debate about co integration vs correlation, as I can see the merits of both, and as some have mentioned co integration is more complicated and a bit harder to understand, however most of us here are using correlation and it seems to be working fine. Can someone here please show an example of a highly correlated pair which has low co integration? would be interesting to look at.

Neither do I.... because I don't see the subject on what we should debate: I haven't seen any academic paper or research stating that the ratio of two correlated series is stationary and - as a consequence - is mean reverting; but if someone have one's, please let me know... I'm glad to learn something every day.


As for an example of a correlated pair with low cointegration, the already cited Ernest Chan in his book talks about Coke and Pepsi (p.130-131) but I suppose we can keep on talking forever if they are highly correlated or not (correlation can change a lot in a just few days).


In the same pages he is also very clear on the differences by correlation and cointegration, when he states " having a positive correlation... it doesn't guarantee that the stock prices will not grow farther and farther apart in the long run even if they do move in the same direction most days".


Ok, that's was just a break.
Please keep on with your journal.... I'will not disturb any more.


(*) by the way, I'm not endorsing Chan's book: it doesn't tell anything special, but it does has one PLUS compared to its peers: he speaks a language anyone can understand, without giving up almost nothing of the "technical jargon".

 

[tatankas]

Quote from jonnysharp:

I plugged this pair into pt finder, and I tell you what, you have found a great pair to trade, Im going to keep it on my radar, it has a historical 100% win rate with a avg win of $1368 per $10,000 invested and avg trade length of 7.33 days and a wide ranging ratio chart makes this pair a good one. If you have a fundamental bias to short ebay long ostk you may want to wait until the ratio goes to 1.50 or above to initiate a position, the current ratio is 1.43 (ebay stock price is 1.43 times of ostk stock price), 150 day high is 1.54, 150 day low is 1.09, avg is about 1.30, see attached.

Technically this pair is perfect. One of the best pairs, i have seen so far.
It's now 2.09 stddev away, so time to enter the trade.
But with such a market capitalization discrepancies and taking in consideration that the OSTK beta is twice ebay's beta, i am afraid we are losing the market neutrality...

I am curious to see, what you will do

 

[tatankas]

Quote from saico:

I think the trade quality of pairs always changes. Therefore I think a shorter backtesting period (12 months) has more weight than a wider timeframe. For myself I update my list about every 3 months.

Yep, i agree on that, we shouldn't forget our average holdin period is around 7 days.

What you think about the following backtest indicator:

Imagine today is 2009 january the first:

X=backtest from 1-Jully-2008 until 31-December-2008
y=backtest from 1-Jan-2008 until 30-June-2008
w=backtest from 1-Jan-2007 until 31-December-2007

backtest_score=0.5*X+0.25*Y+0.25*W

From you personal experience, it makes sense?

 

[Dr Who]

Closed a few trades today which means I've now closed 19 trades, 6 of which have been losers (31%) for a total profit of £472. I have 2 trades open (from yesterday) for a nett loss of £6.30.
All to £1/pt while I'm going through my 30-day trial.
And I'm now quite pleased and have modified my filters for opening trades which has led to an increase in my winning percentage. But of course we're dealing with small numbers here, although its looking promising.

 

[renton]

Quote from Dr Who:

Closed a few trades today which means I've now closed 19 trades, 6 of which have been losers (31%) for a total profit of £472. I have 2 trades open (from yesterday) for a nett loss of £6.30.
All to £1/pt while I'm going through my 30-day trial.
And I'm now quite pleased and have modified my filters for opening trades which has led to an increase in my winning percentage. But of course we're dealing with small numbers here, although its looking promising.

Hi Dr. Who - What are your filters that you run through to decide on whether to open the trading pair or not?