Practical capital allocation

Discussion in 'Automated Trading' started by abattia, Jun 23, 2010.

  1. I have a small number of automated intraday ETF/stock trading strategies that I use each day. Each strategy generally takes between 0 – 3 trades daily. Typical holding periods are anything between 5 minutes and 5 hours. I trade these systems though a retail brokerage account in which intraday leverage is restricted to 4x.

    I don’t have a system or formal set of rules yet for allocating capital optimally between these strategies, but want to change this. I seek your advice and suggestions for practical approaches ...

    I‘m aware of the Kelly formula, and ready to roll my up sleeves and deepen my understanding of how it applies to my situation ... if, that is, you tell me that it’s indeed the best route to go ...

    Is it?

    Or are there better, or otherwise more practical, approaches?

    [If this topic has already been dealt with extensively on ET, please just point me in the right direction, and forgive my ignorance ...]

    As always, thanks in advance ...
  2. Corey


    You want to minimize the signal error covariance between the systems -- i.e. allocate your capital such that when one system is making 'bad' trades, others are unlikely to do so. So if you have 5 systems and 3 all tend to make 'good' / 'bad' trades together, by optimizing your allocation on signal error covariance, you will end up treating your five as a basket of 3 and 2 other (i.e. 3 'overall' systems), which will diversify your income.

    Another way to do this is do a PCA on the return correlation or covariance matrix. This will split your systems into 'orthogonal' groups. Treat these groups as one system when allocating capital.

    But if you don't have the 'math' skills to model that sort of thing, I would just do a basic correlation analysis of their returns. Systems with highly correlated returns should be treated as one system, and capital should be allocated to them as a group.
  3. Thank you for the guidance.

    A basic correlation analysis of the system returns seems like a sensible place to start.

    (... and quite fiddly enough, too, as I am going to have lots of fun with Excel just making sure the data for each system aligns correctly by date with that for each other system!)

  4. Read these:

    before risking one dollar.
  5. Thanks!