BioComp Dakota with Swamp Technology

Discussion in 'Trading Software' started by chifai2, May 19, 2006.

Thread Status:
Not open for further replies.
  1. chifai2

    chifai2

    SupermanTrades likes this.
  2. Humpy

    Humpy

    I think you may mean swarm technology.
    I bought the program on the basis that it will do less curve fitting in its optimisation. Some of the the bots do show a profit on the S&P 500 but its not very big. I am hoping they will write some better ones. I have bought a book on Visual Basic to try to understand what they are saying but regretably I am no programmer. I think it has possibilities..............:(
     
  3. I have it. It walks thru data adapting as it goes and it shows you this paper trading. So I think what you see is more like real.

    I made some systems using bots they give and I'm paper trading them. I want to see what happens before actually thinking about trading with it.

    The systems I made, some like the videos, some my own, 11 have made money (85%) and 2 lost money since I made them. My worst loser is like their MCHP video, it is down about 3% in 4 months :( . My best system is up more than 10% in 3 months. :) If I convert all the gains to annual returns and simple average them (no portfolio mgmt), it is like 20% or so but it is early yet to tell. Have to watch more and make more.
     
  4. slacker

    slacker

    Ok, this looks like a vendor spam with 3 new posters who just happen to be interested in the same product at the same time and happen to visit this site at the same time.

    But hey, up late and bored so here goes.

    This looks like a genetic algorithm that is designed to produce a better backfit of data. It would be interesting to see how the backfitted results of Dakota would compare to backfitted results of a simple one-dimension backfit of system on WealthLab or Metastock.

    I personally like Genetic Algorithms. I especially like to watch them 'learn'. The Dakota program looks like it is only using bots with very few number of parameters that can be optimized. At then end of the run the results would not be much different than if you ran the same type of backtesting several times using Metastock's optimizer.

    Dakota 'might' be interesting if you could run bots on several markets at the same time. Or on a single market using different time frames. It doesn't do either.

    Dakota may make it easier to provide a backfitted results with a few clicks of the mouse. However, anyone want to tell me how they were surprised with the optimized parameters that were the result of one of the runs? Is this approach good enough so there is some 'discovery' of a new combination of parameters that were not considered without this swarm approach. I don't think so....

    It looks way overpriced, and is targeted toward new users who do not yet understand how easy it is to produce a backfitted equity curve on any market. You do not need 'swarm', GA or Neural nets to produce these screens or these results.

    But hey, just my opinion.

    Welcome to ET guys!
     
  5. Slacker,

    Funny thing. You are exactly backwards on what the product does.
     
  6. C_Cook

    C_Cook

    Hi Guys,

    Thank you for the post on BioComp Dakota, which has been brought to my attention. I'm the CTO of BioComp. I'll try to make this non-commercial, but just clarify the technology.

    Specifically to "slacker"s post, Dakota does not use genetic algorithms and is specifically designed to not back fit data. Instead, Dakota "walks-forward" adapting trading system parameters bar-by-bar, showing you "out-of-sample" trading results. This adaptation is done using "Swarm Technology" or other algorithms, which determine new parameter values each bar based on equity performance over the last so many bars. Back-fitting is one of the challenges with most trading tools where back-fitting does not typically perform as well going forward. Thus it would not be proper to compare Dakota's results to backfitted results of WealthLab or Metastock or other trading tools of that nature. It would be appropriate to compare Dakota's trading results to the results by those other tools on data that came after any backfitting by them.

    Also, Dakota uses trading bots with any number of parameters that can be adapted. In the videos, there are bots with 1, 2 and 4 different parameters shown.

    Humpy and Max: Since it sounds like you are not trading yet, please log into the Dakota/Profit forums and read the thread on "How do YOU trade with Dakota". One guy in there has posted how he does it on SP emini's, in detail, including how he setup Dakota and the points he has pocketed.

    Thank you again.

    Carl
     
  7. Hi Carl,

    Your product looks interesting but I don't understand the real difference between backfitting and walking forward. Isn't walking forward adaptation same as constantly optimize (backfit) the parameters? Thank you in advance.

    The second question is how can I get in your beta testing program for Dakota RT? :D

    Regards,
    KC
     
  8. C_Cook

    C_Cook

    Thank you for your question. The distinction is subtle yet important. Back fitting usually means that you look at some large length of history and find *the* (a single set of) parameters of the trading system then use those values going forward. The tighter you fit those parameters to provide seemingly good equity gains and risk on history, the more the system's equity and risk performance tends to decline in the future. Many refer to this as "over fitting", "curve fitting" or similar terms.

    Walk forward adaptation means that each bar, given recent performance (like over the last 50 or 100 bars), you *move* the parameters towards better performance. How do you know where better performance is? One of two ways presently: 1) you know what the parameters and performance were and what they are now and you walk up the the resulting equity slope using a built in proportional, integral and derivative function (aka "Equity Controller"). 2) You run multiple systems with a bit different parameters and then each system sees the parameters and performance of the others and they move their parameters towards better performers and repel away from poor performers (aka "Flocking"). Another way to look at this is having 25 charts with the same trading system open, but the charts talk to each other and adjust themselves towards the better performance each day and you play the Top N charts, the average signal from the charts or just the best chart. Your choice.


    Dakota R/T, the intraday version, is currently in beta testing and we have already enough users helping find the "opportunities for correction". Beta testing is going well and we will be releasing Dakota R/T hopefully very soon, before the end of the month.

    Thanks! I'll be happy to answer any more questions.

    Carl Cook
    BioComp Systems, Inc.
     
  9. Hi Carl,

    In no way I am trying to put it down, the product looks fun and creative. Maybe I don't quite understand the subtleties because I still got the impression that if I optimize my system parameters every N trades, I could probably get the same effect of Dakota?

    The other question is how "often" does the swamp technology able to improve system performance because it seems to me only natural to have some scenerios where it could make the performance worse?? And why?

    Thanks again,
    KC
     
  10. C_Cook

    C_Cook

    Not likely, and here's why:

    First, you would need to readjust every bar and not optimize, but move towards improved performance. Often times optimal is exactly NOT. Performance is transient and often is in a region of system parameter values. This is a topic of discussion in our forums today: Why is the "best bot" not the best. One of the conclusions is that the "optimal" performance today may not be the what you want to trade tomorrow, that "the best" may actually revert to the mean on you and be a poor performer in the near term.

    Second, reoptimizing can cause the trading system to change substantially, causing discontinuities in your signals, which can prevent the system from fully developing the trades and delivering the equity "as traded" rather than "as seen" in the reoptimized trading system. It is better to shift the parameters of a trading system gradually, rather than reoptimizing and jumping.

    Third, you would need to create about 25 or 50 or 1000 charts and retune each on each bar and then play the average or Top N. This would be hard. This parallel averaging inherent in Dakota is automatic and helps.


    Very good question about the Swarm technology :). Dakota enables you to test this easily as you can turn off or change or constrain the adaptation strategies and set the number of bots = 1 so it runs more like a "traditional" tool. We ran some tests comparing a fixed system with N-runs of adaptive systems for a number of tickers and bot types, automatically exporting the statistics to disk. Doing some statistical analysis, we found in the majority of cases it helped, in a minority of cases was neutral and in a few cases it degraded performance. I don't recall any degradations outside of statistical tolerances, that is to say fixed performance was not more than 1 or 2 sigma from the mean of adaptive. It depends on the ticker, the bot (trading system) and perhaps other parameters. It is possible to chase equity and not catch it.

    So, you can test this and use a system in Dakota fixed or adaptive. Your choice, but the more fixed the higher the risk, I think. The natural tendency is to squeeze the equity curve to high straightness and gains "in-sample" (back testing to minimize risk and maximize return) and that comes at great risk of pushing that risk into the future, when real money is on the table.

    Besides "fun and creative" as you say, (we do get those comments often), Dakota offers a different point of view, developed with a unique strategy, to consider when money is on the table.

    Thanks,

    Carl Cook
    BioComp Systems, Inc.
    http://www.biocompsystems.com
     
    #10     Jun 23, 2006
Thread Status:
Not open for further replies.