You are using an out of date browser. It may not display this or other websites correctly.
You should upgrade or use an alternative browser.
You should upgrade or use an alternative browser.
Pair Trading Strategy Journal
- Thread starter jonnysharp
- Start date
Quote from total_keops:
Yobo, thanks for the comments. I highly appreciate. This system is return to the mean but I plan on building a scanner for dispersion. The trade you describe on the SPY:TLT make sense to me and I had already planned to play something around that.
For this specific trade, BWX:IEF, I dont see them as always inversely correlated. Correlation is time varying. 30 days lookback attached.
I always thought that adding a divergence allocation to a pairs portfolio could reduce the variance of the PnL and therefore increase the sharpe. Is this something that you observe? I should backtest it but I have like 25 other systems that I am working on right now and this one is not in priority.
KEOPS,
your correct about the correlation not always being negative. In the short term you will see correlation moves up and down the ladder converging and diverging. This is why I look at a longer time frame to smooths things out. The one year correlation is negative and that is a more meaningful number. I am also using a 100 day moving average to set the allocation and directional change for the pair in order to shed the day to day noise.
Here is the clencher, if you also track the 20, 50 as well as the 100 day moving averages you can adjust allocation of capital accordingly to reduce volitility and draw downs. For example, looking at the spy:tlt pair, once you know the correct order of the pair based on whether it trades above or below the 100 day average, you can adjust allocations based on other moving averages such as the 20 day and 50 day day.
Here's how I do it. Spy:tlt breaks above the 100 day average go long spy 80% long tlt 20%. If its a true move upward, the pair ratio will continue higher. When the move is over, you will first see the pair break below the 20 day average. If it closes below the 20 day, readjust allocation to 70/30. If the pair continues to break down and goes below the 50 day average, readjust allocation to 60/40. If it breaks below the 100 day average flip the allocation 20% spy and 80%TLT. And hopefully you will catch the big move down making money and reallocating on the way back up 30/70 to 40/60 to flip 80/20.
By trading this method you are always on the right side of the market trading with investor asset preference, equities versus treasuries/bonds. This is the beauty of asset diversification.
The end result is that you have reduced volitility and are maximizing returns through tactical asset allocation. This is not a day trade and not to be confused with correlated pair trading. It is a strategy based on modern portfolio theory and the capital asset pricing model.
Here's one more caveat and a way for locking in profits. Use the 100 day two standard deviation bolinger bands. When the pair gets stretched outside the bands simply take money off the table. For example, you start with $100,000 and you got the direction of the pair right and it trades outside the bollinger band and the market value is now worth $115,000. Take 15,000 off the table while maintaining your allocation. Doing this is important if you incorporate leverage, but I wouldn't use more than 2-4x leverage. Otherwise you might get spooked and emotion will overtake the trade. Just remember to keep your eyes on the moving averages and the outer bolinger bands for profit taking and reallocating to reduce draw downs.
As of todays close, the spy:tlt ratio is below the 50 day average and above the 100 day average. The trade is 60% spy, 40% tlt. If it goes below the 100 day flip allocation to 20/80 and if it goes above the 50 reallocate 70/30 and so on...
Hope this makes sense. Good luck.
I too have started looking at co-integration instead of correlation pair trading, but one question I have is once you have calibrated the co-integrating coefficient, when and how often do you re-calibrate the coefficient?Quote from total_keops:
I used about 110 days so close to 6 months. A small sample in fact. It may not be cointegrated over one year but over 6 months it looks like it is, unless my code is wrong and it's possible. If you look at the chart that I attached you can see that it is trading in a range. But I agree that over one year it is not as good. This is the human filter after that that decide to get in or not. I dont qualify pairs and then only look at them. I look at all combinations and if I like it when it comes I place the trade. My approach to the thing.
I used such a small sample because the ETF in my scanner that has the smallest number of days is that and I normalise all my data to it. Something I have to improve :-(
Got out at a spread of 2.17 (BWX @ 57.06 and IEF @ 90.45). Lost confidence in the move. Small profit.
Yobo, thanks for your reply. It does make sense to me.
SilverRanger, a better question is how far back you look! This is a tricky one IMO. But I recalculate the ratio everytime, it's all automated. It should not change significantly over a small period of time like say 3 months if you look back over 1 year. But after 6 monhts it can change. It's a time varying thing like correlation. I dont really have an answer to that.
SilverRanger, a better question is how far back you look! This is a tricky one IMO. But I recalculate the ratio everytime, it's all automated. It should not change significantly over a small period of time like say 3 months if you look back over 1 year. But after 6 monhts it can change. It's a time varying thing like correlation. I dont really have an answer to that.
The look-back period is really a question of short-term stability. Normally, I try to look back 10x the distance I am predicting stability forward. i.e. if I want my pairs trade to last 2 weeks, I expect strong co-integration over the last 20. However, I have been toying around with making this factor more 'adaptive' -- i.e. the 'crazier' markets get, the longer longer I require my period to be relative to how long I will hold forward (as 'crazier' markets tend to imply that stable relationships are more likely to break down). How to really achieve this goal, I haven't perfected...
Am interested to know how others approach a pair where both have been impacted by negative news(DO missed earnings est. RDC - Analyst downgrade).
I viewed the pair as 'news neutral' and took the trade as it was at Layer 2 (as defined by my settings in PTF) and % -> Mean was near historic levels.
I viewed the pair as 'news neutral' and took the trade as it was at Layer 2 (as defined by my settings in PTF) and % -> Mean was near historic levels.
Am going to post selected trades to forum. Objective is to inform and be informed.
Sold GG @ 34.43
Bought GFI @ 11.15
Trade taken intra day. Do not wait for MOC if I like signal
What I like about trade:
At Layer 2(2.5 SDs as defined by PTF)
% -> Mean at near term historical levels
High corr. and trend is not pointing down
No earnings or news driving the pair.
Neg:
Spread chart is opposite of where I want it to be, i.e. would like it to be at 50d high as I am short the spread.
Sold GG @ 34.43
Bought GFI @ 11.15
Trade taken intra day. Do not wait for MOC if I like signal
What I like about trade:
At Layer 2(2.5 SDs as defined by PTF)
% -> Mean at near term historical levels
High corr. and trend is not pointing down
No earnings or news driving the pair.
Neg:
Spread chart is opposite of where I want it to be, i.e. would like it to be at 50d high as I am short the spread.