Mechanical vs Neural trading systems

Discussion in 'Automated Trading' started by pandabear, Nov 11, 2007.

  1. Being a software engineer by profession and an avid trader, I believe automated trader is the holy grail.

    Considering this I have been working on trading systems for a while but haven't been able to create a good intraday trading system (mechanical).

    So recently I am considering a neural system would be a better approach to take.

    So I guess what I want to ask is the following:

    1. Do consistently profitable mechanical trading systems exists? (I am sure they do but does anyone here run one?)
    2. What more profitable for the longer term? Mechanical or a neural system?

    Finally, I think I am in grave need of mentor ship from someone who has or does develop a mechanical/neural trading system on a ongoing basis. What I offer in return is someone whose passionate about the financial markets and someone whose known to offer a different but valid point of view every know and then.

    Thanks
     
  2. mechanical is better - i just know it is cant tell you why - cause i dont want to set here and type that much. mb :D

    anyway you are at the fork, pick the wrong way and its a long haul back, in fact most people i know never make it back.
     
  3. Automated trading is simply a manner of automating a bad strategy, so you can lose money more rapidly. Neural trading is a manner of forming a Rube Goldberg machine that somehow almost never works in the real world. Mechanical trading means nothing unless you have a serious edge.

    Other than that...
     
  4. yea i should have mentioned rule 1. you have to have an edge.
     
  5. Although NNs and nowadays SVMs may form part of a trading system (i.e., classification for security selection, etc), you'll find that the people in the know have woken up and have reduced their dependence on these. If you think that you can just throw OHLCV and these will magically find the holy grail for you then you are very much mistaken.
     
  6. you know one suggestion i will make to you is that i myself fell into the trap of becoming addicted to the chase and not the kill. i still fight my demons every minute, i can often hear my father who has passed saying to me "mark leave it alone its good enough" those hauntings have forged me to stay focused.

    mb
     
  7. Murray Ruggiero

    Murray Ruggiero ET Sponsor

    People look at Neural networks wrong. They are a powerful tool to improve a trading strategy. They should not be the core of the strategy. One example would be developing a system which uses MACD. If you find a profitable and robust strategy you could use a neural network to predict the MACD a few bars in the future and then replace the original one.

    This approach is the best way I have found to use neural networks as a component in a strategy and not the strategy themselves.
     
  8. Hi,

    Some comments to your questions:

    1. Of course they do. I run several, one is being tracked by collective2.com. PM me if you want a link to the system.

    2. How do you distinguish between "Mech" and "Neural"? Neural means many things to many people.... Do you mean a system that isn't trained on in-sample data versus one that is?

    Regards,

    Mike

    P.S. Mods, generally, is it appropriate to post a public link to a collective2.com system here?
     
  9. vikana

    vikana Moderator

    The biggest problem with Neural Networks in general is that they have very large degrees of freedom. In other words, you can train it to produce just about anything on historical data, without it having *any* predictive value.

    I have found that NN and general evolutionary algorithms are very well suited for specialized optimizations, where you e.g. want to explore a huge parameter space to get some sense of where "good" value exits. I have not found working NN systems built just on training a network without a strong underlying model already being present.
     
  10. maxpi

    maxpi

    The most powerful computers are the ones between our ears. They aren't very fast but hey... I recommend developing some ideas and trading them live, maybe with a simulated account, as an adjunct to any other kind of development you do. You will see things and learn things doing that.

    I was able to move from live trading, which I really am not good at or keen on, to semi automation where the computer was a useful helper, now I'm looking a little further down the road to where I can offload more to the computer and free myself up to do the discretionary work with better focus... I can't envision the day when the computer teaches me much of anything or does all the trading, it could backtest some stuff and improve on it, that is the most it can do in it's current state...
     
    #10     Nov 13, 2007