Consistent Profitability

Discussion in 'Automated Trading' started by maxdama, Jul 31, 2008.

  1. maxdama


    I re-read an old post from acrary which stressed the importance of being consistently profitable over a certain interval. It made a lot of sense to me to approach profitability in this way because it's logically the only way you can trade for a living. If you don't get a paycheck each week, you miss your rent or eat less- it's not just some arbitrary number.

    I took an algorithmic approach so that an automated system could hypothetically calculate trading goals for itself. I created a stochastic model of net weekly trading results to find out the required profit factor, winning percentage, and trade frequency to be confident, with statistical certainty, that a system will not lose money.

    At the end of each week you look at your results to see if your system is performing normally or not. Abnormal and normal are based on confidence intervals so it is not an emotional decision to cut a system off after a drawdown.

    Here's a chart of the results, the curves are the minimum profit factor (r), trade frequency (n), and winning percentage (p) to be consistently profitable with 90% certainty:

    You can see that it is crucial to have a decent winning percentage unless your system can generate hundreds of signals to trade on each week. The number of trades required per week is actually exponential with the decreasing winning percentage. It is apparent that it is better to have a system that makes a few good trades per week rather than one that frequently miscalls trends.

    I was wondering how you all determine targets for your system- it seems illogical to rely on a system to remove the emotional factor from trading when the assumptions it's based on are gut feelings. Do you use Monte Carlo VaR analysis instead of variance-covariance? Do you use a different mathematical model?

    Details of my model-derivation thought process are in this word file and this excel table (warning: Math)


    __________________ - The log of my research on and implementation of automated trading strategies
  2. squeeze


    I looked at your website.

    Out of curiosity do you actually trade anything?
  3. maxdama



    Good question- no, I'm a nineteen year old computer science student at UC Berkeley. I'm approaching automated trading from the theoretical AI side, whereas I gather that most people on ET approached it from the day-trading side. Which is why i am looking for outside opinions. Thanks for checking out the site by the way- I hope I can bring something fresh to the table.

    Haha, to get back to the thread topic, I'm guessing you didn't create a mathematical model, squeeze? Maybe it was a little too much of a giveaway about my background. But moving beyond the meta-dialogue, rather than ad-hominem debate I'd like advice and feedback.

  4. Corey


    In my humblest opinion, VaR is a waste of time. I don't care if you have a 99% confidence that you won't reach that point -- the fat tail coupled with that 1% chance might wipe you out.

    Expected Shortfall is a much more realistic measure of risk. Quite simply, you need stable (logarithmic) distributions with fatter tails to make any sense of the markets.

    Expected Shortfall goes well with GARCH(1,1) as a prediction method for variance and a dynamic correlation copula ... instead of the basic mean-variance optimization and static variance-covariance matrix...
  5. I am confused. All you need is a positive profit factor for not losing money. It suffices getting a confidence for that.

    I am not sure you can totally treat trading as a probability and statistics game. Many good traders all they manage is a big winner per year and a few small losers and they do that consistently. Others, have a big very big winner every 2 or 3 years and several losses in between.

    If you are targeting high frequency trading you must make sure you include slippage and commisions as well as other random factors that can distort the results.
  6. squeeze


    I'm a professional quant trader and make my living creating market models.

    There is a large pool of academics that produce a great deal of work but who never actually put any of it into practice. Most of this work lacks applicability to real problems. I could see signs of this in your website.

    My advice, if you can scrape together enough pocket money then start to trade a little. It will provide you with experience that will likely change your thinking and help make your research more focused and relevant.
  7. You don't need consistent profitability in your main system or method. Maximising profits is more important.

    The way people get around the "paying the bills" issue is to either

    i) be employed at a trading institution, thus getting a base salary
    ii) have alternative income sources e.g. passive investment income
    iii) have a second system with reliability of profit (which tend to be short-term systems or daytrading)
    iv) live off capital while waiting for the main system to score home runs

    If a trader needs to make money each week to pay the rent, he has already lost.
  8. Hi, I definitely agreed with squeeze. You can only know what trading is all about when you really do live trade (not on paper). By this does not mean that you should just trade live without knowing or having any system. You will need to believe in a system and trade using it consistently and always learn from your mistake. No Greed nor Fear (trading psychology) :D Easier said than done?? well you can only know it when you trade with real money..
  9. promagma


    I have seen systems lose money for many weeks, hundreds of trades. Then right when you proclaim it dead, it goes on a huge win streak. This is the rule, rather than the exception, when you are actually trading.

    Best approach is multiple non-correlated systems.
  10. maxdama



    I looked up the details of the GARCH model and I agree it's a step up from VaR. If anyone else is interested and not completely familiar with it, here's an introduction short tutorial from NYU. I will test a GARCH model soon.


    I was defining profit factor as avg win/avg loss ratio. So with a PF of 2 and winning percentage of 10% you would lose money even though the PF is positive. We are probably thinking of two different things for profit factor.


    I pm'd you, thanks for the advice.


    You're right, there are definitely other ways to make the baseline income. I'll add one to your list: live in your parents' basement. I like #3 "have a second system with reliability of profit".


    Yes, I definitely appreciate squeeze's encouragement. Your point brings up the inevitable issue I'm facing: you have to do to learn but you have to know before doing. I'm sure it will resolve with time.


    I think multiple, non-correlated systems are the best too. Still, sometimes I'm sure you have to cut one off. Maybe the edge you're exploiting is erased by competitors (it's not unique to trading) and it seems like a good idea to use some metric to determine when a system should be shut off rather than gut feeling.

    It seems like the general sentiment is skepticism of risk models. Maybe because of the current problems caused by widespread misvaluation. I'm shifting my focus away from consistent profitability since it's impractical. It might be a good topic for a bank or fund to develop but less useful for individuals. Thanks for all your responses.


    __________________ - The log of my research on and implementation of automated trading strategies
    #10     Aug 1, 2008