Trading System Lab

Discussion in 'Automated Trading' started by Quantus, Jun 29, 2009.

  1. Quantus



    does anybody here have practical experience with the software

    Trading System Lab

    and likes to share some insights, please?

    Thank you very much!

    Also, what platforms are available apart from


    which automatically generate trading systems, include a variety of statistical research tools, and deliver sound/robust systems?

    I am just curious.. :)

    Many greetings,

  2. Trading System LAb is too expensive for me, it's like 60k down plus 20K per year maintenance. I think the developers target hedge funds only. It is based on a commercially available genetic programing engine. I would really like to test it but it does not appear possible.The developers claim to avoid optimization/fitting/selection/survival bias but I am not sure how they do that. TSL generates code for Tradestation and other platforms.

    The second one uses the indicators in Tradestation to build systems. I have stopped using indicators long ago. I am not sure what type of optimization/fitting is involved there.

    I have a license of the third one, APS Automatic Pattern Search. When I bought it I had doubts if it could do anything useful but it was not too expensive anyway and took a risk. It does however exactly what it is supposed to do as it is advertized on the website. This program has many nice features, like code generation for a number of trading platforms but it is limited to price patterns. You need to know how to use these patterns to avoid selection/survival bias. The program reads only text files and if you use a data vendor that does not generate that format easily you have to go through conversions. This is a limitation.

    The forth one is not available. I have no idea what it does.
  3. Quantus


    Thank you very much for your answer!

    Yes, I also wished to be able testing Trading System Lab.

    Concerning SirTrade, I like fuzzy logic somehow, but only heard about two people using this software. What they told, was also not too convincing. Nevertheless, the developer wrote an interesting book, unfort. still not translated from french to english.

    I used APS myself years ago.. have to admit I didn't get profitable trading systems. My main point of criticism would be.. nearly all systems had very few trades, it was not statistically significant at all. But it's an interesting approach for pattern recognition.-As you wrote you don't use indicators any more, you probably work with patterns mainly!?

    Well, the grail suite was produced years ago from Wouter in Africa.. he really produced many smart tools, using Robert Pardo's book "Design, Testing, and Optimization of Trading Systems". Over years, an agile and active forum developed, with experienced traders and good ideas.. - The whole software package was running in the direction, what Trading System Lab promisses to be able to do now, but for around 1500,- USD. Unfort., Wouter finally sold the package to an african hedge fund, and many people who have tested and contibuted their great ideas, never saw the final working product. Somehow disappointing..

    You will still find a copy of the forum here:

    I especially miss the inspiring forum, something great for all could have been the result.. but that's business.. :)

    Any other products on the market?

    Many greetings,

  4. Hi Quantus,

    I'm also very curious about TSL. Maybe it will become available in the future. It has many statistical tests it seems but I am still skeptical of potential curve-fitting.

    What version of APS did you use? I get many patterns with enough trades to configure short term trading systems with sufficient significance. The point is you don't just use one or two patterns but you have to combine them in an appropriate way. The developer offers some suggestions on how to go about doing this. I have developed my own algorithm for combining patterns using statistical methods. This is my little edge so to speak. It did not make a big profit in 2008 but I was able to end up profitable. I am doing fine this year. I trade futures, forex, ETFs and a few high liquidity stocks.

    There is another program I tried once but I forget its name now. I think it does permutations of build-in indicators but that took very long to get results or even hang up on me several times.

    I think the key with data-mining in general is to have an efficient algorithm that generates results in a reasonable time frame. I could possibly develop my own but I have no time now for software development of that kind. I also think there is not a lot of money to be made in marketing software. There are always issues with piracy anyway.

    I have an idea about a data-mining application and maybe I will look for a partner to develop it together in the future for in-house use only.

  5. It says it test 400k systems an such a massive amount, even with testing against out of sample data its hard to view it IMO as anything more than a massive data snooping generator, especially on one series of market data. With that number of chances I don't see how you can have confidence in any confidence interval.
  6. Valid point.
  7. Quantus


    Thank you very much to you both!

    Concerning APS, I used an older version in 2004. I never received (minimum) 100 trade signals on the historical data, and if you have a look on the company's webpage, even the test pictures from 2008 show mostly 45-60 trades for each system. I really wonder how this should be significant, but I am glad you developed profitable systems/combinations with it. I ordered the latest book of Michael Harris and will give it a try again maybe. - Guess you talked about the p-indicator.

    I am also thankful for the comment about 400k systems in one hour, as I didn't read this. Yes, valid point!

    Judging the equity curve, maybe this link could be interesting for you: (?)

    'regression fitness function - smoothing out the equity curve'

    Many greetings,

  8. APS patterns are what I call micro-patterns. I use them quite effectively but at the right time only. I get patterns each with 150, 200 and some even show close to 300 trades. It all depends on the parameters you specify for the search. If you search crude oil futures historical data with the program, for example, and you demand a profit target of $5 you won't get any patterns or maybe 1 or 2 each with very few trades. If you use only 1000 bars in your search do not also expect to get a high number of trades.

    My suggestion for you involves lowering the win rate to get some patterns with high number of trades. Then increase the win rate and get some more will less number of trades. See when the lower win rate patterns give a trade signal. Do you get a high win rate signal at the same time or right after?

    I stop here..I said enough...
  9. ...complex tools easily drive to complex overfitting... imo
  10. Actually some of us are still using the Grail in differing capacities. Only now am I realizing how far ahead of it's time is was, especially at the retail level.

    Even on the institutional side, you would be surprised how advanced and far along some of the concepts in the Grail are, and how much of an advantage the tight integration the suite as a whole is...genetic optimization, rolling and anchored walk-forward, cluster analysis, portfolio optimization, etc...

    have yet to see anything like it a the retail level, though of course there are genetic optimizers and walk-forward tools, but have yet to see them as well-integrated and as powerful...

    the stress testing on the genetic side is definitely an edge, as is the cluster analysis on the WFO side...

    #10     Jul 2, 2009