investment weighting with signals

Discussion in 'Strategy Development' started by jasonm, Jul 12, 2010.

  1. jasonm


    i am hunting down some literature regarding portfolio/investment weightings with rolling strategies. my example is:

    with this i mean:
    at time t you have wealth W. you have investment weighting x (0<x<=1)

    at time t+1 you have a signal telling you to invest in IBM based on some indicator your monitor. you invest x*W

    you now have a total wealth of .9*W available to invest

    at time t+2 you have a signal telling you to invest in GOOG based on an indicator. you invest x*W, with x*W in IBM

    in theory, you may have more investment opportunities than you have wealth, and some will be more profitable than others.

    how can you go about assuming an optimal rolling portfolio weighting given new entrances and exits occurring intraday, and some trades spanning many days.

    i have currently been pursuing a fixed weighting, but while this is safe it leaves a large portion of wealth un-utilized, and therefore a large amount of foregone gains (also possible losses, but so be it such is the game).
  2. jasonm



    "you now have a total wealth of .9*W available to invest"

    should be

    "you now have a total wealth of (1-x)*W available to invest"

  3. It's really up to your personality/style. FWIW I've tried crazy weighting schemes and they performed maybe 0.05% better which is not significant. Much easier to equal weight.

    Also consider what kindof market your system is good in. If you get 10 buy signals a day in a bull market, and 1 buy signal a week in a bear market, maybe it really is valid to stay 90% cash in a downtrend, or even 100% cash if the system only does well during bull market periods.
  4. jasonm


    thanks for the response. i am going to need to drill down a bit deeper.

    i currently have approx 1,400 signals since late april. i am sure this is not a huge number by any of your standards. it is coded to perform in both bull and bear, so i will need to see what weighting proves most optimal.

    i was thinking some sort of machine learning system, but even that is tricky. signals may prove very good for a while, then die off showing a probability of a high payoff when really it will be small, or negative.

    i am going to need to dig a bit into this.