Best Composite System

Discussion in 'Strategy Building' started by rtstrading, Sep 16, 2003.

  1. Ok, we're talking intraday bar systems for eminis.

    Lets say I have 100 individual "systems", which are all profitable historically, some better than others, but individually do not generate many signals.
    Each system has a maximum stop loss and a target.

    Individually, these are either "buy" or "sell" systems.

    If I combine these systems, only one signal is taken at a time to conclusion, either getting reversed, stopped out, target hit, or EOD close.

    So, what is the best way to determine the most profitable and highest $/trade combination of these systems?
    I realize that this will be a very optimized combination for the historical data set, but I believe the only way to proceed to have the most appealing combination system for future use.

    I currently do it manually combining the most profitable systems to find the best combination, so it is a hit and miss type of thing. All in either Metastock or Tradestation format currently.

    Is there any software that will find the best combination? Neural Nets? or something else? Just curious if anyone else has done this. Brain is locked up, so I'm looking for a key :D

  2. maxpi


    I have never done this but in Tradestation 2000 you can code all of your signals into one strategy with a parameter written into each signal such that the signal can be made operative or inoperative and have the optimization switch them all off and on until you have tested all combinations. The system report can be made to tell you which system caused each trade. 100 systems is going to produce an unbelievably huge number of combinations however if you want to test all combinations up to and including all 100 of them.
  3. T-REX


    kEEP IT SIMPLE.....

    The holy grail is right in front of you.

    Daytrade method ...(for daily profit potential)
    Swingtrade method ......(to catch longer term trends)

    put these 2 together (as I have done) and you have the closest thing to the holy grail you will ever see.

    good luck.:D
  4. balda


    what T-REX is trying to say is this:

    stop loss is 15 points

    profit target 3 points

    I am just kidding :D
  5. I have already contacted the people at Trading Solutions to see if there software would help me achieve my desired results. They said by the end of the year,

    "We are currently adding the ability to TradingSolutions to optimize committees and to optimize neural networks along with their parameters. We hope to have this functionality available before the end of the year."

    The problem is getting my "systems" into TS. I need to rewrite all of them in TS format. I currently have about 800 systems for the NQ, ES, and YM markets, which makes the rewrite a very formidable task.

    I am currently looking at an index of system worth, as follows:

    [Total Profit(for backtest period)]/(Standard error bandwidth) = SW(System Worthiness)

    A system that has an equity curve with a narrow standard error bandwidth is very linear, and is the type of equity curve I want for the composite system. That combined with a high profit system and we have a very good combination for inclusion in the composite system.
    I am currently looking at "Trade Equity" for Metastock which with some work may be able to produce the index value for my systems. With that in hand, I will be able to automatically select those systems with the highest index for the composite system.

    The composite systems that I currently design are by visual selection and are profitable, but I am always looking for steps to more fully automate the process. The systems currently run in automated order entry each day as they are 100% mechanical .

  6. RAY


    If you know excel well enough you can do it there.

    It seems like it would be a a really big optimization model, using solver, and it would take tremendous effort to input it all. Good Luck. :)
  7. Bob, I have absolutely no prior experience or theoretical knowledge of composite strategy trading system design. I also have no problem with making a fool of myself in public, so here's my 2 cents. The closest parallel I can relate to is testing a blackjack strategy. The house rules allow flexibility and decision making both in bet size and playing strategy. For the sake of simplicity, the figures which follow are not accurate but are used for illustrative and comparative (vis-a-vis trading) purposes.

    Playing with incorrect strategy, the player has a 5% disadvantage. Of course this is variable, depending on the extent of his mistakes. If you add in non-optimal strategy and bet sizing, one could easily reach a 10% disadvantage.

    My job is to incrementally reduce the 5% disadvantage by going through each possible playing decision and optimizing it. This could be akin to your 800 systems. (Or not.)

    So for example, if by splitting 8's versus the dealers 6 (as opposed to standing) gains me 0.2%, then I will adopt this "system" and by doing so I am now playing with a 4.8% disadvantage. Similarly, by optimizing and combining the hundreds of other "systems" covering splitting, doubling down, standing, hitting for all possible combinations, it would be possible to reach conditions of a 0.2% disadvantage only.

    Bet-sizing using a count system (which is what you are trying to achieve now) could elevate us to a 1% edge.

    There are certain splits and double downs which, while only increasing EV (expected value) by a marginal amount, contribute significantly to the variance ("standard error bandwidth"). A player might prefer a 0.7% edge with a variance of 8 to a 1% edge with a variance of 24. Similarly, varying the bet size increases variance. An unavoidable fact of life is that this particular edge is accompanied by huge swings in bankroll (relatively speaking).

    I use mathematics or Monte Carlo simulation to test my "systems" and to eventually test my "composite" system. You used backtesting (and perhaps forward testing) to come up with your 800 systems. If all systems are positive EV, then the composite system could be thought of as generating the blackjack count, as it were. For each period in the timeframe, there is a number N (where N is either -800, -799, -798, .... , 0, 1, 2, 3,.... or 800) which is the sum of all 800 system signals (+1 = long, 0 = neutral, -1 = short), ignoring weighting for the moment. Working on raw tick data, I would be able to analyze and optimize this "composite" using primitive software which I would develop for this purpose (Turbo Pascal), much in the same way as I have done with blackjack. There are advantages (many) and disadvantages (few) to using third party testing software like TS, WL etc. You are now up against one of the disadvantages.
  8. T-REX


    I don't understand ALL THAT TECHNICAL STUFF!!!!!

    All I know is that I personnally use EXCEL. THAT'S IT!!!

    No TradeStation, META STOCK, etc.etc.etc.

    I just use what works. All that scientific stuff I just avoid. I'm not knocking anybody who understands them and uses them but I dont see how ALL THAT FUZY LOGIC can help you make money.

    Maybe I'm just not mathmatically inclined or lack the academic ability to comprehend but "Bet sizing" "Optimal F" and anything requireing a Doctorate in Calculus is ABOVE ME! and MOST traders out there who are too ashamed to admit it!!!!!!!

    I don't understand HALF the stuff some of you guys talk about. All I know is what works:

    DayTrade Model
    Swing Trade Model

    i.e. Buy when the maket is trading above the 40-Day MA
    i.e. Sell when the market is trading below the 40-Day MA

    That one simple rule will MAKE YOU RICH!!!!
    Anything in addition to that will make you CRAZY!!!:confused:

  9. Mr Sub,

    I just finished Bringing Down the House by Ben Mezrich. Sounds you might have been one of those MIT blackjack team guys.
  10. AAA,

    I have had my fair share of adventures worldwide, but this was not one of them. I take this opportunity to belatedly clear myself from another of your suspicions - I am not Swing Zones.

    #10     Sep 18, 2003