Need some help about trading systems

Discussion in 'Automated Trading' started by Crashh, May 9, 2005.

  1. Crashh


    Hi ppl, I'm new here, and I'm liking already the quality of this forum.
    I need some help, regarding trading systems and indicators.
    I love programming so I've done dozens of trading systems and mechanisms and indicators in the last months, and I want to let some become public if ppl likes them (for free of course), and I use to find the better indicators by doing some simple one-indicator trading system and back-test it alone in an historic of data (I know over-fitting is bad and not advisable I just optimize them to know how much potential they have).

    Now, the question is: when u do a new indicator/trading system, and test it for ex, in Metastock, which values should I desire at least in each of these results:
    - Winning trades / loosing trades (I've heard somewhere it sould be around 40%/60% W/L at least?
    - Profit/Loss Index?
    - Average win / average loss (better than 2.0?)?
    - Maximum number of consecutive losses?
    - Minimum system close drawdown?
    - Maximum number of system/indicator variables? (5?)
    - Relation between profit in an optimized history (like 2002-2004) and a non-optimized new history (like 2004-2005, simulating this way real time data)? (we all know that a system that does in 2002-2004 would certainly fail at 2004-2005).
    - What % to demand from an optimized past history (even with over-fitting and full optimization)? (Because sometimes I reach (including margin calls) 12 0's numbers (yes I know that's useless because it's optimized for that, and yes I know I shouldn't use ZigZag's :D) My logic is, if an stochastic wouldn't give more than for ex. 1000% in 2 years (optimized to the full) and one indicator could give 100.000, the second would have more potencial on a daily basis.

    And should I (when testing an indicator/trading system performance):
    - Use big Leverages? Or 1/1?
    - Use spread?

    I test them nowadays with an leverage between 10/1 to 50/1 with a margin call of 50% (so I can test well it's potential even with margin calls and big leverages), without take-profits nor margin-calls, and usually start with spread 0 and try later with spread 2, avoiding using more than 5 variables (maximum) on the indicator, and use to search an average w/l of more than 2 and w/l trades of more than 40%. I also try to do indicators with only 2 or 3 parameters so it can be less curve-fitting.

    Testing without spread may seem bad but what spread should I put there if there can be so many? :)

    I have done lots of indicators (specially for forex) using euro-usd hourly (mostly) for back-testing. I have indicators from modified moving averages (but more profitable and less failures), to oscillators, bands and all kinds of stuff and I only pass to the next indicator (and have more ideas) when I get profit on that one.

    I just want to know the values that make an indicator/T. system perfect for giving it to the world (I don't want to give stuff that all ppl have at their homes, I want to contribute with something usefull).

    By the way, is there any entity (maybe "Stocks and commodities magazine"?) that study and accept contributions of our custom indicators and show them to the world?

    You're my only hope ppl because in my country I can't find answers neither in some other places I've tried :p

    Best Regards,

    Gonçalo Nuno
  2. toe


    Honest answer is there are very few performance metrics you can look at alone, they generally need to be seen as a balance between several. One that can be used independently is Sharpe Ratio (Modified Sharpe Ratio above 3), but I would still want confirmation from several other indicators.

    These aren't the only problems to overcome before you can trade a system though. Apart from slippage and commissions you need to think about some way of testing for curve fitting. You need to test Out of Sample on lots of data just to start with.