Are some parameters' values "special"?

Discussion in 'Strategy Development' started by luckyputanski, Aug 5, 2012.

  1. I'm wondering what's your view on possibility that some values for parameters can be special? For example, you trade on hourly chart on 24h market and you find out that setting some parameter's length to 24 gives much better results than average of, let say, 10 - 40h. Or you trade market that's opened only 8 hours and although a parameter's settings 5 - 15h are profitable (and the best is 10h), setting length to 8h gives better results than average of 5-15.
    The same for daily charts. What if length of 260 days gives better results than walk-forward-optimization? Which one would you trade?
    In one way, it sounds like curve-fitting, on the other, if a parameters make "sense" it seems like a good thing.
    What's your opinion?
     
  2. Nym

    Nym

    what you said works quite well in backtesting, however I would be reluctant in apply this on-line...
     
  3. jcl

    jcl

    Special parameters in a backtest can have two reasons:

    - Either they trigger a single trade with a particularly huge win or loss that affects the whole test. That should be easy to determine. It is always recommended to remove the biggest single win and loss out of the backtest result.

    - Or, in the case of time periods, they are in sync with some seasonal effect, such as the NYSE opening or closing hours. An indicator of this is when the special parameters are multiples of 4 or 8 hours. This effect is not curve fitting, but real, and can be used to improve profits.
     
  4. dom993

    dom993

    Change your chart timeframe by +/- 5% ; 10% ; 15% without changing anything to your strategy, and see how the system's performance is impacted.
     
  5. How can I change daily chart by +/- 5%?
    Or do you mean parameter's value?
     
  6. dom993

    dom993

    Convert daily to the number of minutes in the day-session (assuming you are charting day-session only, else do the same using all-sessions), then do +/-5% to the result. You will probably have to create a session template that goes from Monday's open to Friday's close in just 1 session, though.