Not logged in
Viewing 2 posts - 1 through 2 (of 2 total)

Forums>StrategyQuant>General Discussion>Monte Carlo: randomize strategy parameters

  • #111461 |
    70 Posts



    I noticed that some of my strategies built in SQ survive the Robustness checks but only if I do not select ‘Randomize Strategy Parameters’.


    1. What is your opinion about such strategies? Would you place bets on such strategies, or, discard them because they did not survive the Robustness tests with option ‘Randomize Strategy Parameters’ checked ?


    2. If a strategy might be curve fitted for historical data, then, robustness checks by variations in historical data alone should be sufficient…why then variation checks in strategy parameters are also necessary?


    3. Out of all the options that can have variations for Robustness testing, the strategy is in my control whereas other options (historical data, start date, skipped trades etc) are not in my control. I anticipate no Act of God that would suddenly tamper with my strategy variables. Do you? :) Kindly share any info on what options in Robustness checks that you feel are absolutely necessary and sufficient!


    Thank you

    Mark Fric
    1182 Posts



    in my opinion if strategy doesn’t pass Randomize Strategy Parameters test it really suggests that it is curve fitted.

    Robust strategy should work even if you slightly change the input parameters. But what exactly does it mean it failed? How much worse was the drawdown and net profit?


    It depends also on how you developed this strategy. Did you use at least two Out of sample period tests? Then the risk that strategy is fitted is smaller.

    If it passes all other robustness tests, I’d try it also with Walk Forward optimization or WF Matrix. Then you’ll see the effect of periodic reoptimization on the performance.


    It is also a question if it is worth putting so much effort into one strategy that already failed one robustness test.



    StrategyQuant architect

Viewing 2 posts - 1 through 2 (of 2 total)

You must be logged in to reply to this topic.