Entries statistically significant compared to random backtested entries?

Discussion in 'Strategy Building' started by Dhalsim, Jan 11, 2013.

  1. Dhalsim

    Dhalsim

    Trade data (vol filter):
     
    #21     Jan 23, 2013
  2. Dhalsim

    Dhalsim

    No volatility filter:
     
    #22     Jan 23, 2013
  3. Dhalsim

    Dhalsim

    No volatility filter equity curve:
     
    #23     Jan 23, 2013
  4. I personally run away from systems like the plague that don't have at least a 2-to-1 win/loss ratio and around 50% winners give or take a few points...this is my drop dead minimum

    ...and don't say it can't be done because I trade for a living and do it almost everyday...intraday timeframes


    How? trade management to keep losses small and deep researching of market behavior to identify repeatable and consistent price patterns...I don't just throw up a bunch of indicators and see what sticks...you have to look under the hood...get oily and dirty and stinky and really study market behavior...


    ...keep losses small and put yourself in a position for big winners (but I'm not looking for homeruns on every trade...only after I have met my monthly, quarterly, and yearly goals)
     
    #24     Jan 23, 2013
  5. Dhalsim

    Dhalsim

    Thanks for advice. However, i did say i use zero indicators (this is both in my discretionary and automated trading strats).

    As for 2:1 winners to losers on intra-day time frame and especially on ES that is so difficult to do. I have tried my tests on having at-least 2;1 win to loss targets to stops - all systems perform significantly worse for me anyway. On longer term strats this can be done.

    The more i look at it, i feel most days especially low volatility days have pure random data on intra-day basis - i don't see how some people can have similar results year in year out as opportunities on low vol years are limited. Hence, the need for having a portfolio of strats running together.

    Can you give any insight whatsoever onto repeatable patterns without giving away any edge related data. Anything at all system based instead e.g. idea for developing strat instead of analysis of results.

    I would be very grateful if you could shed some light on this.
     
    #25     Jan 23, 2013
  6. The pattern itself is probably not that important as I believe that I could take many different patterns (elliott wave, harmonics, five waves, classic edwards and magees, peter L brandt's patterns, you name it - i'm just pulling these off the top of my head...I don't use any of these) and make them profitable. The key is in trend identification to filter the trades which I can't get into because that's a big part of my edge. However, draw upon your own experience and break the problem down into:

    1. Lightbulb moment generates a trading idea
    2. Research the idea to make sure that it makes sense
    3. Analyse the success of the pattern during a backtest (in-sample, OOS)
    4. Develop enough results to meet your own confidence level (Bill Eckhardt uses a minimum of 1,800 results)
    5. Forward test the idea with virtual money to see whether it will hold up under real world results (This is your test bed)
    6. Start very very very small with your position size.
    7. Only increase your size out of profits. If you stumble, drop your size immediately and go back and start with either step 5 or step 6
    8. rinse and repeat on another pattern

    What did the above cost you financially?....hopefully not much except your time. Plus, if you performed steps 3-5 correctly and thoroughly you should be making money.
     
    #26     Jan 23, 2013
  7. dom993

    dom993

    Dhalsim: You should certainly go ahead and program the Long side ... that shouldn't take you much time, and will provide you a lot more information.

    If I read your reports correctly, you can backtest about 6 years (2007-2012) ... if your strat. needs volatility to outperform, its best results should be 2008 then 2011.

    Now, there is 1 aspect I really don't understand ... you claim it is an intraday strategy ... how can the average trade (win or loss) be ~ $1000, that is 20 ES points ... hard to see that with 1 contract, intraday basis.

    If your strategy scales into trades, I would suggest to backtest each scale-in separately ... You'll have a much better visibility on what contributes to the results.
     
    #27     Jan 23, 2013
  8. Dhalsim

    Dhalsim

    Hey Dom, the system does not scale it is all in and out and sorry but i forgot to add that when i ran the back test the program had more than one contract as size.

    With esignal 11 extended history on 1 min chart i only have access going back to 2007 (which is certainly an issue). The calculations by esignal are strange how they are calculated - the most accurate one is no volatility filter as this is clear representation of what the strat was intended to exploit, looking at that data: 2007,2008 and 2011 and best years by far.
     
    #28     Jan 23, 2013
  9. Dhalsim

    Dhalsim

    Thanks Hurricane.

    I am curious can you put up a screenshot of what one your systems equity curve would look like. Also, can you screenshot data like i have - it would be nice to see what breakdown of a system with 2:1 winners would look like especially year by year results.

    Thanks.
     
    #29     Jan 23, 2013
  10. Sergio77

    Sergio77

    Here's an honest guy. Some who insist it can be done obviously do not trade.

    There is software where you can specify targets and stops and other performance criteria and it will find repeatable patterns for you based on those. In that way you can find out what works in a market and what doesn't. priceactionlab dot com comes to mind. Another one is stratasearch but I haven't used it for a long time.
     
    #30     Jan 27, 2013