Pair Trading Strategy Journal

Hi Johnny,

I downloaded pairtradefinder and went through their list of industries and back tested them. There are just too many stocks, even ones that correlate well.

Do you have any advice on a cutoff for correlation, like 90%+, 95%+, etc., and any experience with which industries seem to trade the best for you?

I need a way to "pair" down the list so to speak. Thanks.
 
Quote from tatankas:

Hi John,

First of all, let me thank you, for all the shared information.

I have read about pairstrading sometime ago, but on that time, i found it unsuitable for individual investors.

Now, with pairtradingfinder, i am changing my mind, but still afraid of the strategy effectiveness, when the market conditions shift for a clear trending line with low volatility.

Anyway, i am doing paper trading, based on ptf signals and some homemade filtering (if you have spare time, take a look on: Short CIEN long GLW).

Regarding the signals generated by ptf, i understand a entry signal is generated when ratio deviates 2STD. Do you know if this is the only data considered? RSI and volatility is also taken in consideration?

What about the exit signals. Do you have any idea how are they calculated?

Best Regards
More...

I don't think pair trading rely's on high volatility to be profitable, actually my pairs went askew with the sharp increase in volatility in early october however when volatility decreased they quickly snapped back. my charts tell me to long CIEN short GLW not the other way around as you suggest? yes the default is 2 stdev, you can set multiple layers at higher stdev's. Im not at liberty to disclose the exact formula methodologies as I only received them as a paid member upon request with condition of non-disclosure. but yeah there not too complex.
 
Quote from LazyLightnin:

Hi Johnny,

I downloaded pairtradefinder and went through their list of industries and back tested them. There are just too many stocks, even ones that correlate well.

Do you have any advice on a cutoff for correlation, like 90%+, 95%+, etc., and any experience with which industries seem to trade the best for you?

I need a way to "pair" down the list so to speak. Thanks.
More...

What I did was enter all the stock codes from the industries with more than 10 stocks in them, backtested them and applied a filter to only show results above 70% avg correlation and avg profit per trade above $300, then i used the ratio chart in the backtester as I do with my entry signals to find the best pairs in each industry, Ive only got about 70 pairs on my radar, each pair is unique, i.e...the same stock isn't in 2 pairs plus no more than 3 pairs from a particular industry. you can use the backtester data view option and export results to excel, in the spreadsheet there is a running PnL column of each pair, you can collate these together and create a portfolio equity chart. historically technology and financials or old economy stocks, ones that sell commodized products so their business's are highly dependent on a central price, i.e..energies, all have been good to trade, but what you want is a diversified portfolio of pairs.
 
First, thanks very much for the thread. I've been interested in pair trading for a while and your thread has helped get me focused.

My problem is getting started with the software.

I downloaded the package. I first tried to put all stocks in by sector, ran the backtester and saved all results with correlation > 70% but my list got so long I crashed the program.

Then I cleaned out the whole database and put in one by one the industry list(s) they supplied for each industry (i skipped financials). Then I once again have been running the backtester and this time have only been grabbing those that come up with correlation > 85%.

I figured once all the pairs got in the main watch list I would try to narrow them down some more form the main list.

Unfortunately I am about 1/2 way through the industries and have about 800 pairs (I am not analyzing them beyond finding 85% correlation) and I am crashing the program again, presumably based on the size of my list (I have sent them an email to diagnose). I have a high powered system w/ lots of memory so it should not be a computer issue.

They have been very good in supporting me but i'm getting frustrated in that i must be doing something wrong in my approach.

So... putting aside that their software shouldn't be crashing (hopefully they can fix the problem) i figure i'm just approaching this the wrong way.

If you don't mind maybe you can give me a little guidance. As I mentioned I run the back tester by industry then just grab all pairs with a correlation > 85%. You have far less pairs so...When you are at the back tester level do you analyze them in more detail before putting onto the list?

You also mentioned that you never put the same stock in twice, but if you have one stock that correlates with 3 others very well how do you choose which one to put it onto the list with? My thinking is put them all on the list and then look at the charts when a signal comes up to choose - maybe at a given time one stock will be correlating with a different one better so why not put them all on the list?

Thanks for the journal and any advice you can supply.

Thanks for any thoughts on this.

Mike
 
Quote from jonnysharp:

What I did was enter all the stock codes from the industries with more than 10 stocks in them, backtested them and applied a filter to only show results above 70% avg correlation and avg profit per trade above $300, then i used the ratio chart in the backtester as I do with my entry signals to find the best pairs in each industry, Ive only got about 70 pairs on my radar, each pair is unique, i.e...the same stock isn't in 2 pairs plus no more than 3 pairs from a particular industry. you can use the backtester data view option and export results to excel, in the spreadsheet there is a running PnL column of each pair, you can collate these together and create a portfolio equity chart. historically technology and financials or old economy stocks, ones that sell commodized products so their business's are highly dependent on a central price, i.e..energies, all have been good to trade, but what you want is a diversified portfolio of pairs.
More...


Thanks Johnny. I have everything over a 90 correlation in the pilot right now. I will start eliminating them as I go to trim the fat.
 
Testing strategies...... Divergences in Full Stochastics seems to be pretty reliable. Just my observation. Take it or leave it.
 
Quote from jonnysharp:

Exited 2 trades;

Sold ETP @ 34.93
Covered PAA @ 37.80

Sold ANV @ 5
Covered GRS @ 6.55

Opened new trade;

Long ZEUS @ 15.87
Short SIM @ 4.60
More...


Jonny,


With your SIM trade at what price did you get short? Was it the 4.60 MOC print (like you said u usually get the trades at the MOC print) t
 
The trading stragegy we have been examining is mean reversion pair trading, dollar neutral. Something I'm beginning to look at, using the data from Jonny's spread sheet is Unilateral Pairs Trading. One assumes that of the two stocks in the pair, one is the cause of the divergence from the mean. If that one can be identified, it can be traded and the other one not. (Yes, that means loosing the protection found with trading the other stock of the pair, the hedge). It is suggested that a way to identify the "wayward" stock is to identify the one with the highest volatility.

I want to test this strategy. I need to know where to find historic volatilities for the single stock of interest (not the pair, which PT gives us). I'll enter that into Jonny's spread sheet and see if there is any correlation between the volatility and profit in each pair Jonny traded since August 2008.

Any suggestion on easily finding historic volatilities? Are there lists, or a charting site where I can simply enter the symbol, date, and get my stat? I've looked without success.

Jonny, thanks for sharing that spread sheet. It's been a terrific learning tool for me. I've analyzed almost every trade to learn how to identify winners from loosers. Still learning.

Walt B
 
Quote from waltbx:

I need to know where to find historic volatilities for the single stock of interest...
Any suggestion on easily finding historic volatilities? Are there lists, or a charting site where I can simply enter the symbol, date, and get my stat? I've looked without success.
More...
If you have the stock data you can use Excel and compute the stdev() on a rolling window of x days. A simple as that.
 
You must log in or register to reply here.
Back
Top