Proper Paradigm For Successful System Development

Discussion in 'Strategy Development' started by PinkBunny, Sep 21, 2006.

  1. This post is aimed at those who have succesfully developed mechanical trading systems. I have just started, about two weeks ago, spending my days using relatively sophisticated proprietary software to develop a mechanical system. I'm having fun and learning a lot about what doesn't work as well as realizing the extent to which markets are random - I know they're random with islands of inefficiency/exploitable fractals, but all I'm seeing is the noise.

    I'm not interested in nor would I expect anyone to share their system with me. Even if you did, it wouldn't do me good because when it dies I would be dependent on it and unwilling to cut it off; I also would have no way of finding a new system. I want to learn this the hard way, and if that means spending six hours per day for two years banging my head against the wall trying to get a system, I'll do it.

    I'm interested in hearing someone succesfull describe their development methodology - what goes on in your mind, where do your ideas originate (from watching markets? Do you just think of things that should work?) and how much time do you dedicate to one idea? Do you keep building a system that doesn't work to see if you can get something out of it, or do you scrap ideas quickly and focus on simple systems? Is simple more effective, or are all basic ideas discounted by the markets and is the only chance at exploiting a consistent edge to be found through a complex, unique algorithm? Is there anything systematic about your system development, or do you just try things until you find something that works? How do you avoid wasting your time on bad ideas, and how do you know you're spending too much time exploring the wrong path?

    Answers to any of those questions would be highly appreciated (or being told that I'm asking the wrong questions and approaching this problem from the wrong angle). There's a huge amount of vagueness to this, so I guess what I'm asking is: if you were a newcomer and starting your process all over again, what non-system-specific information would you want to know to expedite your journey to success?
  2. es175


    there's a poster by the name of acrary on ET. unfortunately for people like you and me, he's left the site for pastures new. on the bright side, his posts are still here. highly recommended reading.

    his journal on system development, linked below, might answer a few of your questions.

    good luck.

  3. Prevail

    Prevail Guest

    become an expert in one market. master daily data first, then intraday. go to to the forums, read chuck's 50 or so papers and see the historical progression of system design. then spend 10 hours a day for 2 years finding out what might work.
  4. Think data.
    Think virtual physics.

  5. Give your entry a reason.

    Read Acrary's posts.
  6. mspkash1


    I've been trying to develop some systems for more than a year.
    The systems that do work all have one thing in common: They are simple. You don't need some sophisticated indicators to make you money. Remember all indicators only follow the price. You want some indicators that can predict the future price.

    Also if you are manually developing systems and back testing one system after another, it'll take forever to come up with something. You'd want to use tools like stratasearch to find you a good system. Then improvise on what you find.
  7. Dustin


    I would first ask yourself what type of trading you are interested in. Do you like high volume scalping, or low volume trend following or breakout type stuff?

    Then look at the daily movers and see which stocks would have been good for that kind of system today. Then see if you can find the edge in trading that strategy which is the hard part. For example if you like breakouts then find the 2 or 3 stocks that worked today. Then look at what the stocks had in common...were they liquid or low volume? Once they broke out did they retrace? When was the optimal exit point...or did they run until the close? Did they go with or against the market?

    Was today a big volume day relative to normal?

    If so, could you have scanned for it before the breakout, and in that case could you have found a common entry that allowed you to accumulate stock at a low risk entry point (meaning tight stop)? If the breakout was worth on average .7% and your stop loss would have been .2% could you make a system out of that?

    This isn't mumbo jumbo. I recently had to redesign my entire trading and went through this same process and quickly found a fairly profitable system.

    BTW I am a non-believer in backtesting although i'm sure it can be done. I think the best type of testing is live trading with small lots.

    Good Luck
  8. Instead of trying to find the optimal system for the market, focus on Identifying the type of market you are in and apply the system that works best for that particular market. Believe it or not, simple systems do work. It's just a matter of doing alittle bit of work, and learning from the market and your trades.
  9. saxon


    I would say that an indespensible part of developing a trading system is WALK-FORWARD testing.

    Backtesting is fine just to see if you are in the ball park. But it is too prone to curve fitting to be relied on as a final test of your system before trading it live...especially if your system employs some kind of optimization of parameters.
  10. I suspect there are only a few general categories of trading systems. One is trend following. Another is pattern recognition. Another might be reversion to mean. I write computer programs to test trading methods. I find lots of methods test profitably.

    Following are the results of one of my systems. This system is based on the idea that a moving average has a slope. Some people might think of it as a rate of change of the moving average. When the price of a security shows a low price then rallys the moving average value decreases then increases. Here I compare the value of the 50 day exponential moving average with the value 100 days earlier. When the new value becomes greater than the 100 day earlier value then I buy. When the new value becomes less than the 100 day earlier value then I sell. This test uses about 10 years of Intel Inc. symbol INTC daily price data. Risk is pretty high. The system is not optimized. Experimenting with different values helps understand how the system performs.


    EMATimeConstant is 50
    Slope is 100
    ATR_Multiplier set to 10.00
    Name of data file is INTCdail.txt
    Long or short is: Long only
    Heat is 10000
    Initial capital is 100000
    YearsOfData is 10.00
    Slippage and commission per transaction is 100.00


    24-Jul-96 OHLC:[ 8.30 8.78 8.25 8.63 ] buy 8.83 size 2215
    Dollar Position Cost is 19558.45
    6-Jan-98 OHLC:[ 18.47 18.58 18.17 18.28 ] sell 18.19
    Position Net Gain Or Loss is 20732
    Subtotal profit $ 20732

    21-Aug-98 OHLC:[ 21.14 21.42 20.59 21.24 ] buy 21.30 size 890
    Dollar Position Cost is 18957.00
    19-Oct-00 OHLC:[ 42.02 42.31 40.88 41.94 ] sell 41.63
    Position Net Gain Or Loss is 18094
    Subtotal profit $ 38826

    10-Jan-02 OHLC:[ 35.43 35.47 34.50 34.65 ] buy 34.82 size 591
    Dollar Position Cost is 20578.62
    3-Jun-02 OHLC:[ 27.43 27.72 26.44 26.62 ] sell 26.23
    Position Net Gain Or Loss is -5077
    Subtotal profit $ 33749

    29-May-03 OHLC:[ 19.80 20.97 19.78 20.83 ] buy 20.87 size 1205
    Dollar Position Cost is 25148.35
    22-Apr-04 OHLC:[ 25.97 26.75 25.85 26.52 ] sell 26.80
    Position Net Gain Or Loss is 7146
    Subtotal profit $ 40895

    4-Mar-05 OHLC:[ 24.70 25.01 24.49 24.68 ] buy 24.68 size 1236
    Dollar Position Cost is 30504.48
    7-Nov-05 OHLC:[ 24.08 24.55 23.95 24.50 ] sell 24.38
    Position Net Gain Or Loss is -371
    Subtotal profit $ 40524

    Number of trades 5
    Total profit $ 40524
    Profit after subtracting $ 100 commission & slippage per transaction: $ 39524
    Cumulative Annual Growth Rate (CAGR) is 3.95 per cent.
    InformationRatio is 0.81
    Heat is $ 10000 per trade
    Drawdown is 0.0366
    Initial capital is $ 100000
    Slippage and commission per transaction is $ 100.00
    Uses 20 period average true range to calculate position size
    This system calculates long position trades only


    This system at it's peak only uses $ 30504.48 out of the initial capital of $ 100,000. Approximately 70 % of funds are available for other trading. Either I can increase risk to make full use of capital or I can add other securities to the portfolio. I write computer programs to test these possibilities.
    #10     Sep 23, 2006