Algorithmic Trading Gets Smarter After Quant Upset

Discussion in 'Wall St. News' started by archon, Oct 4, 2007.

  1. atonix

    atonix

    Exactly. The Black Swan: The Impact of the Highly Improbable covers this very well. This man has made much from managing risk of these "impossible" events.
     
    #41     Oct 7, 2007
  2. i'm not much of a statistics guy, but the probabilities used to determine likelyhood of events based on standard deviation seem flawed. It makes sense in financial markets that the size of sigma itself increases quickly as abberations occur. So what is denoted as a 10x standard deviation event in typical speak is more likely as being much more highly probably than that. Its probabilities of occurence are more similar to something with much fewer std deviations.. i guess sigma and standard deviations should probably be determined on a logarithmic plane, to determine more realistic occurence probabilities in financial markets.
     
    #42     Oct 7, 2007
  3. It's actually not that difficult to figure out when the markets go into "panic" mode. If I can do it, I'm sure these quants can.
    I went flat from the end of July until the start of Sept, pretty much, based on my own look at standard deviations. To oversimplify, a part of my system depends on trading when a stock goes more than two std's away from its "normal" trend (but defining normal is itself an artform, of course).
    When I see multiple financial stocks fall out of this two std range, and not come back in, so that trades that are normally not only profitable but my most profitable wind up losing me money across many stocks at the same time, I know the markets have entered panic mode. And you can't model panic, by definition. These quants seem to think they can, for some reason.
     
    #43     Oct 7, 2007
  4. panzerman

    panzerman

    Std. Dev. of what? Price, volatility, or some other measure? The problem is that the probabilistic nature of the markets is not modeled very well using Gaussian statistics. Fractal statistics are still a new enough field that nobody has developed a good market model yet with them.
     
    #44     Oct 7, 2007
  5. Sure you can... maybe not using classical statistics, i.e. the standard gaussian theme, but, markets are elastic and herd behavior is fairly reliable. There are plenty of well researched non-linear approximators that accurately (with a known degree of error) model market panics.

    The major flaw with correlation studies, especially when used for dynamic hedging, is the assumption that the correlation will continue for an indefinite amount of time, and, that any deviations will "snap back" eventually. The result being no defined "uncle point"... IMO this is a no-stop-loss type approach, hence the blow-ups we saw this summer.
     
    #45     Oct 7, 2007
  6. The point is to avoid getting hurt in a liquidity crunch.

    Whether the SD of the event is 25 or 4.3 or 3.7...
    Is of absolutely no value.
    Mathematics and models will not help you much in extreme markets.

    In fact...
    Quantitative analysis is of relatively little value in extreme markets...
    And people who stick to it blow up...
    It's always the same story...
    "I stuck to my model... and blew up".

    ONLY 10+ years of daily trading through several market cycles...
    Will give you the wisdom to not only avoid getting crushed...
    But to actually EXPLOIT extreme markets.

    Do 1,000,000 trades over 10 years...
    And you will understand what I am saying.

    As a corollary...
    This is exactly why Pro Traders crush Black Boxes in extreme markets...
    WHEN IT COUNTS...
    Because human intelligence is orders of magnitude ahead of primitive Black Box AI.
     
    #46     Oct 8, 2007
  7. nitro

    nitro

    Aahhahahhaahah. :D

    I would love to put my model running on a computer against you or any human trader on the planet on a fast moving market with fear rampant. You would get obliterated.

    The reason these models lost money is not on positions they took when the market went wild, it was on the positions they already had on when the markets went wild. Once the volatility is out of the bag, the models know what to do.

    Volatility regime prediction is impossible. It is why Taleb threw millions of dollars down the drain waiting for it, and why Neiderhoffer makes millions for years, then blows up one day. Two sides of the volatility risk coin.

    nitro
     
    #47     Oct 8, 2007
  8. maxpi

    maxpi

    recognizing the correlation takes largely some computer work maybe, recognizing when the correlated items have decoupled is another story...
     
    #48     Oct 8, 2007
  9. STD of the log of the price. You can fairly reliably model a stock most of the time with this, if you're careful about it. However, human intelligence has to be applied when extremes start to come in, or you wind up losing lots of money.
    I don't believe there is such a thing as a perfect model, so I don't try to make mine perfect. I just have a few rules for recognizing when it breaks down. Because in real life, sh*t breaks.
     
    #49     Oct 8, 2007
  10. vansmarket

    vansmarket Guest

    you can't squeeze blood from a rock.

    that is what these quant hedge funds are trying to do.

    no real volume or not enough..

    plus the size of these funds are' too big' to trade in illiquid markets.





     
    #50     Oct 8, 2007