Pairs Trading Strategy Model

Discussion in 'Strategy Development' started by Neutral_Al, Sep 5, 2002.

  1. Hi WDGann,
    The ideal pairs trading system would incorporate some elements of the FA, especially the quantifiable ones. It would be much better for your good night sleep to be short a "bad" stock and "long" a good one. However, if you prioritize FA over TA and correlation, you end up with long/short portfolio. I'd like to base my model on correlation simply because the idea that the 2 stocks of choice have been doing the same thing for a significant period of time gives me comfort.
    If you are looking for a book, I just got my Pairsrading manual in the mail. Haven't read it completely yet, but it looks like they have all the formulas I need. Go to to check it out.
    Hope this helps,
    #21     Sep 17, 2002

  2. Hi CHrisM,

    Correct. However, you are not interested in one particular correlation ex: ge/ msft. You trade 10, 20 at a time. One stops working, so you take a stop loss. The idea is that statistically you have higher chance of a profitable trade based on correlation. People at claim 80% and show the record. I still have to verify that for myself. The numbers though could be different because I am still not sure about entry/exit points.

    Will keep you posted,
    #22     Sep 17, 2002
  3. ChrisM



    I trade myself also T-Bonds, and their correlation to inflation figures make sense. But what`s the idea of ge/msft (except trading in one boat related to broad index) ? I know that market is a living organism and has no respect to any logic, however, can create things which can vanish in seconds.
    But You`re right about statistics - that`s all about.
    I`ll try this (ge/msft) myself on our machines. I`ll let You know what I find.
    #23     Sep 17, 2002
  4. ChrisM,
    I used ge/msft as a general example. Actually, the correlation between these 2 stocks is only 28% according to my calculations. The stock that correlates with GE the best is TMPW (87%). It is interesting that AOL correlates with GE @86%. If you run it on you machines, let me know if you get same results.

    #24     Sep 18, 2002
  5. Hi Dave,
    I use end of the day data from TC2000 right now. It takes forever and it is all manual. I need an automatic source with real time data stream so I can try to attempt to build a daytrading system. Any thoughts?
    Here is a pair that I think is an ideal trading vehicle for 2-7 days trades: FO/ITT. Look at the daily ratio chart with 40DMA. I opened
    a position today at 13:50 long ITT short FO R=0.836.
    Let me know what you think.
    #25     Sep 18, 2002
  6. al..... there are better stocks to pair GE agains ... there just not one to one ratio.... i trade ge against Honeywell... or tyco... dont trade it against aol you will get least stay in the same sector or news will kill you... tyc and ge has been good on a 2:1
    #26     Sep 18, 2002
  7. royce09


    May I ask how you found that pair? Seems very interesting?

    Thanks Royce
    #27     Sep 18, 2002
  8. DaveN


    Hi Mark,

    Thanks for the reference to the books! I'm looking forward to reading yours. (It's on its way from I enjoyed your TASC article as well.

    I'll post some follow up responses and plan to add to this thread with some thoughts, ideas, or discussion about your book.
    #28     Sep 19, 2002
  9. DaveN


    Hi Al,

    As you can see, I'm just catching up on this thread. Sorry for the delayed reply. While Tradestation has its liimitations, as Mark Conway pointed out in the TASC article, I have done most of my systems work in EasyLanguage. I think that Wealth Lab, Neoticker, and AmiBroker might be helpful in your systems, as most will interface directly with eSignal or and will manage two data streams or more. I don't use these packages, so I'm not clear on the exact benefits of each.

    What you ultimately select will depend on your programming propensity. I am very interested in comparing EOD correlations with 60 minute, 30 minute, etc. to determine whether different timeframes can help me be more selective based on the types of pairs arbitrage systems I want to run. i.e. a shorter, faster trading system with lower average trades is probably better selected using a shorter time compression for your correlation runs.

    If you know Visual Basic, or C, or some other programming language, it's possible to run studies on nxn spaces in fairly short periods of time. That is, to run the correlation of each of, say, 600 stocks against each other. I plan to do this in Visual Basic for Applications using Excel. I'll build my system directly in Excel, and not only run the correlations, but also record the system results for each pairing. I'll be looking at that resulting data set in many ways. This may not be the direction for everyone though.

    I'll take a look at that pair that you suggest. Thanks for the info!
    #29     Sep 19, 2002
  10. Hi Royce,
    I found this pair by calculating 2 year correlation and adding some strict parameters on 40DMA over 6 months. The book that I bought at helped a lot as well.
    I understand your concern about FO / ITT . One is a furniture business and the other one is a waste manager. I was surprised myself. But think about it, if these two companies have been doing the same thing for 2 years, there is about 90% probability they would do it again (based on my S&P500 testing so far). Plus they all are within S&P500 which imposes certain correlation as well.
    Now, if you want a pair within the same industry, how about Fannie Mae and Freddie Mac (FNM/FRE). They look so good to me that I am opening a position tomorrow at the open.
    FO/ITT is making money so far. Will see what happens tomorrow.
    #30     Sep 19, 2002