News algorithms?

Discussion in 'Automated Trading' started by dancingphil, Feb 3, 2006.

  1. Hi there,

    I've been hearing a fair bit about algorithms that can supposedly read news and trade automatically off it. Sounds pie in the sky to me, but has anyone come across this??

    Thanks
     
  2. acrary

    acrary

    The hottest new category of trading algo's is using immunologic response. It allows the models to become self-adapting and completely autonomous tradebots. Within 5 years I expect these bot's to be better than any human trader ala "Big Blue" in Chess.
     
  3. H2O

    H2O

    Any articles on this?
     
  4. MAESTRO

    MAESTRO

    We have been toying with the Latent Semantic Analysis for two years now. The results are very encouraging. We now can process the real time news and find a definite bias in it. I too think that these kind of developments are the future for the algorithmic trading

    http://lsa.colorado.edu/papers/dp1.LSAintro.pdf
     
  5. I wonder how you would handle the risks implied.
    If you don't really trust your programs you're vulnerable to someone triggering your risk management.
    If you do trust them, well, what was the account size you wanted to manage?
     
  6. This is the kind of wild-eyed talk one has been hearing from the AI community for 30 years.

    Chess is a game of ** complete information **...
    Therefore beating Grandmasters was purely a matter of sufficient computer power...
    And computers became fast enough about 5 years ago.

    To extrapolate chess to the financial markets is CRAZY.
    Today's adaptive AI techiques cannot even come close to beating poker pros consistently.

    Academics with extremely limited trading experience...
    That write theoretical papers on beating financial markets...
    Have been a dime a dozen for generations...
    And their work is mostly a waste of time to read.

    rm+

    :cool: :cool: :cool:
     
  7. I agree with you Redman, the human mind in trading will never be replaced. As long as humans are irrational, so to will be markets. Ask LTCM, they had it all "figured out.":)
     
  8. How can someone learn this? I want to get started in this type of trading so can someone point me in the right direction.

    -Kastro
     
  9. acrary

    acrary

    Outside of investment banking I don't think there's much in the public domain. Immunologic came on the radar about 10 years ago when Mortgage Backed Securities needed a better method of estimating prepayment rates. Someone I know had a idea of trying dynamic estimation based on bio markers and immune response. Lots of work was done and the firm got some Cray's to do the number crunching. It proved much better at estimating prepayments and much more research has been going on since then. What's new now is the desktop pc's are now powerful enough to do the crunching and realtime immune response identification is now a reality.

    I have a model called Doron based on this that is coming up with trades for the ES market using 3 min. data. By itself it's profitable in 85% of the months since I've been running it in realtime. The most impressive thing is the average monthly return is greater than the std. deviation of returns (pretty unusual). It is so compute intensive that only one agent can run per-system. I now have 3 agents running on separate systems and coordinating their trades so that if one takes a trade the others skip it. They then look for another opportunity.

    I'll be starting a paltalk chatroom later this year and have someone post the trades in realtime for all interested to see.
    Once my life settles down in a few months I'll post or write some articles to give a overview of the technology.
     
    #10     Mar 14, 2006