Trading System Expectation

Discussion in 'Automated Trading' started by BenChi, Feb 1, 2006.

  1. BenChi



    I am just learning to program Automated systems with TradeStation and was wondering what should my expectations be for automated system profitability.

    I guess the question is, in an average year, what kind of (percent)return would one expect to achieve with an average automated system?

    Thanks for your time!

  2. Your question is far too broad to answer. System returns, automated or not, are all over the board, and depend on experience, capital, skills, resources, etc. etc. etc.

    Just keep in mind that a highly skilled money manager with considerable capital and resources would be happy to do 20% a year, and most would be pleased simply to consistently beat the S&P on a long-term basis.
  3. Your answer lies somewhere between Positive and negative infinity. Seriously now. There is no "average" automated system. What you need to do is test your systems trades on out-of-sample data.

    For example. say you have 2 years of historical data. Train your system with only the 1st year of data. Then test your systems profit/loss on the second year of data. You must be careful not to test the system on the same data you trained with because your resutls are going to be totally incorrect and biased. Basically you'd just be curve fitting the, you need to extrapolate, not curve fit.

    Depending on the number of trades your system makes you can calculate a probability distribution.. if you only make a few trades then you cannot make very accurate assumptions about future returns, but if you have a very large number of observations you can fit a distribution to that, estimate tail indices (for extreme loss/gain probabilities), skewness, kurtosis, etc.

    Ideally your distribution should be skewed towards the the right and have more returns in the right tail then the left.

  4. It depends only on the guy that's programming.
    You may find this disappointing, but there's no better answer.
  5. DrChaos


    You should learn how to fool yourself, all to make sure that before you put in real money you don't fool yourself.

    I would look at a few issues: "profit factor", gross profits divided by gross losses on back tests.
    Too low or too high is bad. Why too high? (e.g. 3.0+) because then that means you either overfitted or accidentally are using future data: too good to be true.

    And then per trade expectation.

    On the per trade expectation I would subtract out half again of transaction costs, and then maybe multpily by 1/2 to 2/3rds to get "real life" expectation.

    Then with those parameters, recompute profit factor. Pessimistly adjusted PF greater than 1.3-1.4, not bad.
  6. i will usually take my system and apply two years back and my net profit has to be at least 10 times larger than my largest drawdown in that two year period.

    and then the next and most important thing is to be honest with your self and challenge whether you would be able to sit and watch and experience that drawdown live.