Discussion in 'Trading' started by nutmeg, Feb 17, 2012.

  1. Below is a post from Tyler Durden. Somewhere on ET a while ago I had started a thread on this very same subject (I have no clue where it is) (no I don't want a cookie).I really didn't have anything to back up my thoughts on this but Tyler does.

    It appears that while we were busy over the past month spreading the Greek pre- and post-bankruptcy balance sheet, and otherwise torturing Excel (something we urge other financial journalists to try once in a while - go ahead, it doesn't bite. In fact, it is almost as friendly as your favorite Powerpoint) our peer at such reputable financial publications as Forbes, and many others, were laying of carbon-based reporters and replacing them with... robots. As Mediabistro reports, "Forbes has joined a group of 30 publishers using Narrative Science software to write computer-generated stories. Here’s more about the program, used in one corner of Forbes‘ website: "“Narrative Science has developed a technology solution that creates rich narrative content from data. Narratives are seamlessly created from structured data sources and can be fully customized to fit a customer’s voice, style and tone. Stories are created in multiple formats, including long form stories, headlines, Tweets and industry reports with graphical visualizations.”

    " In other words, with well over 70% of stock trading now done by robots, we have gotten to a point where robots write headlines and stories read, reacted to and traded by robots.

    Surely, what can possibly go wrong. And here we were this morning, wondering why the market is not only broken but plain dumb.

    Forbes is not alone:

    The New York Times revealed last year that trade publisher Hanley Wood and sports journalism site The Big Ten Network also use the tool. In all, 30 clients use the software–but Narrative Science did not disclose the complete client list.

    What do you think? The Narrative Science technology could potentially impact many corners of the writing trade. The company has a long list of stories they can computerize: sports stories, financial reports, real estate analyses, local community content, polling & elections, advertising campaign summaries sales & operations reports and market research.

    And here is a sample of robot generated "literature"

    While company shares have dropped 17.2% over the last three months to close at $13.72 on February 15, 2012, Barnes & Noble (BKS) is hoping it can break the slide with solid third quarter results when it releases its earnings on Tuesday, February 21, 2012.

    What to Expect: The Wall Street consensus is $1.01 per share, up 1% from a year ago when Barnes & Noble reported earnings of $1 per share.

    The consensus estimate is down from three months ago when it was $1.42, but is unchanged over the past month. Analysts are projecting a loss of $1.09 per share for the fiscal year.

    Condolences to all financial journalists. If you thought your meager salary was crap, you are about to be replaced by a costless algorithm. The market you wrote about no longer needs you. But at least we will now have computers telling us all about how (seasonally adjusted) trends in financial journalism employment are improving.

    Probably what is even sadder is that nobody noticed as more and more robots have taken over for humans.

    In other news, this development naturally, this adds a layer of variability in market dynamics which make some completely unpredictable feedback loop imminent:

    because if a robot is reacting to a headline written by itself (and it is only a matter of time before Narrative Science is acquired by GETCO or some other HFT behemoth in the latest market manipulation scheme) the epic collapse possibilities are simply stupefying.
  2. I liked the experiment that described Jack Hershey's writing style as a four variable markov chain.

    The prose output from inputs like "walls of the dom" or "ftt" was pretty funny.

    I doubt the same program generating these stories is being tracked by cqg/bloomberg/reuters/newswire computers that can be programmed to instantly read data and transmit orders based on whether they were "better than expected" or "worse than expected" or like where jobs decreased by greater than 12,000 then buy 5,000 ES.

    That's about the extent to which words are used, the rest is quantitative. The programs for prose narrative journalism are almost also certainly used by CNBC, MSNBC, AP, and NBC, but I don't think will ever be the same program reading what it wrote prior to making trading decisions. It's limited to the above.
  3. I didn't finish your long post.

    what's your point?
  4. Good question...

    at the moment I don't have a point..just putting the info out there..but I do know that at some point in time...this will have relevance.

    Reading through the comments on zh, seems like others don't have much insight into impact or effect on trading.

    That's okay though..

    someone is going to exploit this to their advantage.

    Due time.
  5. No, but you pointlessly and mindlessly quoted and reposted it verbatim, which even a robot can be taught not to do.

    The point?

    All machines can and will be gamed.