How (and why) does trading cointegration work?

Posted by on May 22, 2012 in ArbMaker News! | 10 Comments

Many explanations of how and why cointegration works, and what makes it a powerful tool, serve the uninitiated poorly. Forests of algebra do not make for approachability.

Worse, the links between cointegration and tradability are often assumed to be part of a reader’s knowledge – and that is when it is not being rashly implied that cointegration = tradablility and is thus enough to transact upon.

So we have put together this Quick Guide: Principles & Practicalities  to sketch out in simple terms the mechanics behind good potential trades. It focuses on situations where the pre-requisite of cointegration has already been established by ArbMakerPlus and helps answer the question “Now what?”

10 Comments

  1. guan
    May 25, 2012

    Hello admin,

    is it possible to search more than 2 instruments that are cointegrated together to create baskets?

    e.g create a basket of indices like S&P, DAX, Nikkei and HangSeng or a basket of currencies like EU, GY and AU etc.

    best regards.

    Reply
    • Admin
      May 25, 2012

      Hi – we do not offer this today but we are in the preliminary stages of looking at adding it.

      Reply
  2. corporate
    June 4, 2012

    Hi admin,

    1. this platform use ratio model, or spread model?? what is better for %Win?
    2. have you any good strategy for pair´s testing?

    Thx
    Joe

    Reply
    • Admin
      June 4, 2012

      Hi Joe,

      1) We use the Engle Granger cointegration methodology and focus on residuals analysis. The Quick Guide: Principles & Practicalites on this post helps explain – it is a form of spread modelling.

      2) Yes, we provide a sample of strategies which can be downloaded from the Take the Trial page (see point 5 on that page).

      Regards,

      ArbMaker Info

      Reply
  3. Virtual money
    June 6, 2012

    Referring to pages 4 to 5,
    How does ArbMaker analyse/forecast the “residuals” since they are random/normally distributed?
    It is generally mean reverting but how do you predict what is the next value/range between the upper/lower limits?

    Reply
    • Admin
      June 6, 2012

      Hi VM,

      Unfortunately residuals are not always – or even usually – normally distributed in equity and forex cointegration studies. It can be established that mean reversion and cointegration are present but if the distribution is non-normal it is tough to establish the probabilities on when a return to the mean will occur.

      ArbMakerPlus therefore scans for cointegration but includes filters to return those cointegrated pairs that have normal or near normal residuals distributions. All normal distributions have well-known probability properties (as described on that page 5 you refer to) that make it possible to assess the likelihood that extreme readings will be followed by a trend back towards the mean.

      In terms of timing that return, a useful guide is the number and the distribution through time of the axis crossings the residuals chart has shown under the period of study. We provide both the number of crossings and a filter to select those crossing most evenly distributed through time.

      So in sum what ArbMakerPlus does is scan very large numbers of pair combinations first for cointegration; and then picks out those that are most tradable as a result of normal or near-normal residual distribution. By doing so the software takes advantage of the probabiilty properties of normal distributions that are central to statistical analysis.

      Regards,

      ArbMaker Info

      Reply
  4. Virtual money
    June 7, 2012

    Hi admin,

    Thanks for the reply. I was wondering can CoIntegration/residual mean reversion concept(arbmaker software) be applied to forecast this Gaussian distribution time series?
    http://math.stackexchange.com/questions/148585/lottery-sum-forecasting#comment344033_149163

    Reply
    • Admin
      June 8, 2012

      Hi VM,

      Intriguing. It is not clear the observations are independent. If the numbers, once drawn, are not replaced the observations are not independent and identically distributed (IID). IID is a key assumption of normal distributions. So there is some doubt ArbMakerPlus can sucessfully expand into the lottery business!

      ArbMaker Info

      Reply
      • Virtual money
        June 8, 2012

        Hi admin,

        lottery draws (& sum) are independent & I.D.D .
        http://www.catchalotto.co.uk/post/sum-it-up

        It would be nice if you can show an example how ArbMaker can analyse any user input time series data like normally distributed residuals:
        http://sg.myfreepost.com/sgTOTO_analysispower.php?draws=60&fn0=notselected&fn1=notselected&fn2=Sum

        Reply
        • Admin
          June 11, 2012

          Hi VM.

          ArbMakerPlus is not designed for user-inputted series, only financial market data feeds.

          We are setting up a forum on the site soon; that may be a better place to discuss the statistics of lottery results and similarities/differences with financial market data. Two important points of difference: lottery results cannot be arbitraged; and the data sample sizes are small. But certainly there are some interesting academic papers on them.

          Regards,

          ArbMaker Info

          Reply

Leave a Reply