What type of drawdowns do you go through with your system

Discussion in 'Automated Trading' started by trackstar, Jan 28, 2009.

  1. Corelio

    Corelio

    Strategy performance and robustness has ABSOLUTELY nothing to do with how fast you can double your money.
     
    #11     Jan 29, 2009
  2. What kind of money management do you suggest to improve these stats?

    Also, win / loss stats by themselves are not necessarily meaningful. Trade expectancy is what influences long-term viability. If you have 99 trades that yield $10 / contract, it only takes one considerable loss to wipeout a considerable percentage of the gains made over the 99 trades and put the trader in a nasty DD.

    rt
     
    #12     Jan 29, 2009
  3. Corelio

    Corelio

    You can improve win/loss percentage, but that does not necessarily imply lower DDs.

    Same for win/loss ratio. In general optimized systems based on high PF ratios are not sustainable in the long run (not robust). In fact this can be simply verified by optimizing inputs based on win/loss ratios and then testing the input variables on out-of-sample data. You will find that for the majority of the time the system is not robust because the optimization is being done on the price noise rather than the fundamental signal you are trying to capture with your system.

    Noob? Sure, I'm the noob...lol
     
    #13     Jan 29, 2009
  4. Corelio

    Corelio

    You cannot make a system work by just adding "creative" money management if the system is not robust in the first place.

    :p
     
    #14     Jan 29, 2009
  5. I would love for you to elaborate.....leaving me with lots of questions....oh the wonderful world of trading.
     
    #15     Jan 29, 2009
  6. You'd be hard pressed to convince anyone that any market-system won't produce drawdowns, or even periods of flatness (no return).

    More or less, the equity curve the OP has posted does not really have any glaring drawdowns, just periods of relative flatness. Money management, which seeks to optmize geometric returns, won't filter out periods of no-growth.

    But again, I urge the OP to perform a walk-forward analysis to see how the performance can be improved.

    rt
     
    #16     Jan 29, 2009
  7. I built a software tool in C++ that analyzes session PNL data from file. The file contains a scan of the user's parameter combinations, and each combination has a vector of session PNL numbers for the entire window of historical data analyzed.

    The tool breaks up the vectors into smaller pieces to perform a walk-forward analysis, all under the assumption that the user re-optimizes parameters repeatedly at fixed time intervals (i.e. every 10 sessions), using a window history of specified length.

    you can see that the equity return curve can vary by fitness metric used to optimize. What you don't see is how much the curves vary when in-sample and out-of-sample lengths are changed.

    I never look at the tradestation performance statistics, as they are woefully inadequate in judging a system. IMO, system trading requires this level of analysis to maximize the chances that the system under question will perform into the future.

    rt
     
    #17     Jan 29, 2009
  8. here is the second pic
     
    #18     Jan 29, 2009
  9. Very cool to say the least! Needless to say I am not at a point in time where building something of that nature would be possible(lack of education).
     
    #19     Jan 29, 2009
  10. #20     Jan 29, 2009