System made 124% in 1/2 a year but ...

Discussion in 'Trading' started by zanek, Mar 3, 2013.

  1. zanek

    zanek

    So my new system made 124% in half a year, but it has 49% losing trades, which I suspect will be higher once I backtest a few years.

    My question is, would it matter to you all if a system had 80% losers if it was still making large gains overall ? Do you think it matters psychologically to take many losses in a row ? Most of my losers are clustered together so I'm kind of worried about that.

    The longest losing streak was 6 trades with a 15% drawdown. It also had the longest winning streak of 6 wins for a 40% runup.

    Also, for all the flamers , I'm fully aware backtesting in the past doesnt guarantee future results, and that my backtest time period is small, etc

    Some stats from my system:

    Total Losses: 49 (48.51%)
    Total wins: 48 (47.52%)
    Gain % of profit: 124.44
    Avg Loser: ($485.20)
    Avg Winner: $1,195.31
     
  2. I would strongly suggest backtesting for several years.

    I had a one year version of an algo that I created over 10+ years of price data. It beat the 10 year version slightly over the backtest range, but failed the out of sample while the experienced bot kept on slinging the dollarz into the account.

    80% loss rate means high risk of huge DD's and potential catastrophic failure, unless you have an equity curve stop (which you should have anyway, IMO).
     
  3. Why are you asking here? the answer is so obvious. obviously you don't trust your own results so what you should do is obvious: you err on the side of caution and start over.

     
  4. WoodyK

    WoodyK


    Obviously you know or should know a very large sample is required. That's what all the Quants do.
     
  5. dom993

    dom993

    100 trades for a backtest is nothing - really. The only good news here, is that about 5 years should give you ~ 1000 trades - then, you start to get a feel for the validity of the strategy.

    That said, markets change, so no matter what, you'll have to know the difference between a normal drawdown and one that signals that particular edge is gone.
     
  6. Handle123

    Handle123

    For day trading, I backtest 10-12 years of tick data so I have a min of 3000 sample size but often it can be 25,000 signals. The more information you have, better idea you know drawdowns.

    If it is daily bars, I have used 10-30 years.

    Have you tested by using the "Mean" ave of loses? If you notice than the ave losing streak is 5 loses, wait till you have that, then start taking signals. Find the ave.mean of winning signals to stop taking trades. Might really increase percentages.
     
  7. You have not stated if you can back test results for multiple years. As noted the market may be in a bull market for the last 1/2 a year, and your system may not work in a sideways or bear market.

    You also need to understand why your edge works. I know why my edge works and is used by some of the top 1% of traders in the world who have used it with real money for
    many years to become extremely rich.
     
  8. tobbe

    tobbe

    How do you know that it's used by the top 1%?
     
  9. Mtrader

    Mtrader

    I know nothing about trading, like most people here. :)
    But wouldn't it be logic that you have more than 50 % winning trades? If not it is better to gamble because you will have 50% chance to win too.
    I think the first thing to do is to understand the market. If you understand the market you will have more than 50% of winning trades. if you are not able to reach the 50% mark it is clear to me that you don't understand the market and should improve your system.
     
  10. d08

    d08

    If I have a choice of 2 systems with equal Sharpe and Net Profit, one having 49% winners and the other one having 90%, I will pick the one with 49% every time. Percent winners has almost no meaning in reviewing performance but very high win rate systems tend to have hidden fat tails. Trading isn't about being right but about making the most profit with the least risk.
     
    #10     Mar 11, 2013