Successful Automation?

Discussion in 'Automated Trading' started by Norm, Dec 29, 2005.

  1. Norm



    I would like to hear from people who have developed and/or used an automated trading system profitably. There is no need to describe the system (unless you want to - and then you can certainly leave out any critical details). I would just like to know how prevalent such systems are.

    I have read considerably on strategy automation and followed some of the posts here. I get the distinct impression that it is very difficult, if not impossible, to develop an automated system that is consistently profitable.

    Rather, it seems that some systems work well in certain market conditions (systems that are long biased work well in uptrending markets, for example). Other systems may work well for a short period of time and then deteriorate as market behavior changes.

    Anyway, anyone's personal experiences with profitable automation would certainly be educational for us all.

    Thanks in advance,
  2. I am aware of at least 2-3 people on ET alone that runs somewhat sustainable automated trading systems. As I have said before, I run a couple of completely self-built automated trading systems. However, the amount of resources (both time and investment) required to build a successful automated system is significantly higher than a discretionary based system. So far, I have kept track of some 10-12 automated trading systems other than my own, including a couple of truely profitable (at a level of return that just boggles my mind, and I have seen some huge returns, easily > 100% annually).

    However, the "Barrier of Entry", if you will, of a good automated trading system is high, and that's stopping a lot of people without the right combination of knowledge (strategy, back testing, system building, programming, etc) from building their own.

    Just my $0.02
  3. That's about it.
    Automated or not, most posters don't have any experience with profitability - i.e. they are commonly called 'losers'. Remember the popular but never undisputed 95-99 percent losers tale.

    Talking about your "Barrier of Entry", this barrier is already very high for traders, discretionary or system.
    Adding automation to any process of any complexity, trading or not, constitutes an additional formidable "Barrier of Entry".
    If you can handle both, you may get there.
    Most computer hard/soft whippersnappers don't have the spunk to handle this. Hence they are doomed to fail. Not pleasant, but true.

    "Making money while you sleep" will remain a sweet dream for most.

  4. Norm


    Thanks for your input Rufus. As always, it adds much needed substance and sanity.

    I would like to know how complex the profitable programs generally tend to be. That is, roughly, how many lines or pages of EasyLanguage (assuming they can be coded in EasyLanguage) are typically required.

    Also, if you don't mind telling, I would like to know if the profitable programs tend to be trend following or counter trend programs.

    I am presently developing three programs myself. They vary from six lines long to about a page and a half. I believe that each of these is properly classified as counter trending. They are different approaches that attempt to mimic my discretionary trading.

    I have been very successful with discretionary trading over the last two years. This includes 42 profitable trades in a row where the trades were done under a range of different market conditions. The annualize return is between 50% and 100%. It could be much higher. It just depends upon how you annualize, because assumptions must be about about how often the trading opportunities present themselves. Regardless, I approximately doubled the amount of money in my account in about a year and a half of trading.

    I have been trading part time and it is not clear how well the discretionary system would perform if operated full time. That is, if there was an attempt to take advantage of every trading opportunity. Indeed, attempting to take advantage of every trading opportunity may be counter productive. It may be that my discretionary trading - either a conscious or unconscious level - was more selective that an automated system can be. That is, I may have been using entry/exit criteria that I am not even fully aware of. In any event, I may have been using entry/exit criteria that are difficult or impossible to code.

    I am working on three mean reversion systems. This is what worked very well for me in manual trading and I am trying to duplicate the results using TradeStation. I have only been working on the automated systems for a few weeks. The results are mixed so far. I have some encouraging backtests, but I have to scrutinize them carefully to make sure that they are truly meaningful.

    For example, last night my net profit looked astounding at first glance. Then a closer look at the chart showed the the intrabar price movement assumption that TradeStation uses was giving me false results. I was using one year of 1-minute bars. My buys were too frequently executing at the very bottom of the bar. I repeated the test with 30 days of tick data and confirmed that the astounding results were unrealistic.

    Other backtests look promising, without any apparent problems. But again, I am still learning how to properly evaluate the results.

    Anyway, if you could let me know how long profitable programs typically are and if they tend to be of a particular type (trend following, breakout, or whatever), I would certainly appreciate it.


  5. Unfortunately comparing automated systems is a bit like comparing apples and oranges. Most of the automated systems that I know are highly complex systems. these systems are all custom written (in C/C++, Java, etc), and are not dependent on a standard environment (such as TradeStation). To give you some idea of size, some of the systems have 2-3 MBs of source code, of which, 80-90% are custom computation, logic, risk evaluation, etc, and only 10-15% are the main body. The main body of code (the driver and logic) are around 50k - 200k in size.

    Almost all automated systems I have seen are mean reversion systems (I consider arbs, spreads, mm are all variants of mean reversion strategy), since these systems are designed to capture small deviations in movements.
  6. Hi Rufus,

    Can you explain what you mean by "I have kept track of". How do you keep track of them, is there a resource I'm not aware of? or - I think we've discussed that you used to code systems for others - so is that what you mean?
    ie. as a developer you were responsible to babysit your implementation of someone else's strategy?
  7. I have a circle of associates (mostly people I went to school with, and former colleagues), all of whom are involved in automated trading in one way or another (prop desks at ibanks, DMA, hedge funds, etc), so we share information fairly openly amongst the group, I just happen to be the main organizer most of the time for drinks, dinner, etc. We respect each other's need for confidentiality, so you won't find any of us talk about the systems in any specifics with other people. I have no problem talking about general categories, theoretical approaches. Also, not that we designed that way, but the systems don't exactly compete with each other (different approaches), we have found that talking about specific market behaviors (not strategies!) can enhance return for other people's system as well.

    Babysitting an automated system is a huge pain. In most of the large organizations, those guys have hired entry level assistants (junor phds or mbas tend to be popular) to babysit them. My system is still a bit new (a little over a year), but eventually I will probably hire an assistant (automated system clerk / runner?) to babysit my systems. I am the strategiest and developer, and that's what I enjoy doing.
  8. Thanks for clarifying.
  9. So far, 1257 lines of VB/C code, on top of API. Not much but this is my first venture into automated trading. Easy language pales in comparison to what you can develop. The first focus has been building in the risk management modules. I have a development background and have enjoyed every minute of coding it. There is nothing like loading your own front-end system vs broker front-end.

    Dream it. Build it.

    Edited to add: I have manual systems that I have been trading for a while but too complex for easylanguage alone - I wanted to branch some of these off - also have projects that take me away from the trading desk for more than a few days, automation was next step.
  10. maxpi


    I just did a backtest of a countertrend system that required about 500 lines of EasyLanguage. I have it reporting stuff to Excel and doing all sorts of things to characterize the trades for me. The realtime code to implement same might wind up being 100 lines. By far, most of the lines do not run on every tick. On a tick that triggers nothing, as little as 8 to 10 lines of code might get run, if a tick triggers an event then the number of lines of code actually ran on that tick might be 10 to 20.
    #10     Dec 30, 2005