We, Robots: UAL Hit By Perfect Storm Where Nobody Is at the Switch

By: Ray Pellecchia

File Under: NYSE

• 8 Comments
• Email Entry

This is a post-script thought on the UAL debacle I discussed here yesterday. Today I was reading some of the news articles further dissecting what happened, and one in particular caused me to see this event in a new perspective:

UAL Story Blame Is Placed on Computer
Events Remain Murky, But Automated Search, Trades Played Roles (WSJ.com)
As Tribune Co. and Google Inc. pointed fingers at each other over the glitch that cratered UAL Corp.'s stock Monday, blame spread to the computers that robotically troll the Web for news stories and execute stock trades automatically.

Skimming the main facts presented by the Journal:

Google traces the appearance of the 2002 article in its search engine to a process that began late last Saturday night. At 10:36 p.m. PDT, Google's "crawler" -- the technology that finds Web pages -- discovered a new link on the Web site of Tribune's South Florida Sun-Sentinel newspaper in a section called "Popular Stories: Business." The article -- which didn't carry a date but was published by the Chicago Tribune in December 2002 -- hadn't appeared there when Google's crawler last visited the page at 10:17 p.m., the company said.

It remains unclear how the old story rocketed onto the list of most popular stories. ...

From the Sun-Sentinel site, the article became available through Google News service, accessible if a user searched for keywords like "United Airlines." The article didn't appear in any of the headlines on Google News's home page, but it was picked up and sent via email to people who had created a custom Google News alert about UAL or related topics.

The stock market opened Monday with no drop in UAL shares, but the UAL story began circulating widely via a posting by research firm Income Securities Advisors Inc. that was made available to users of Bloomberg L.P., the financial-news service widely watched on Wall Street. Shortly after a headline from the outdated report flashed across Bloomberg screens at about 10:45 a.m., UAL shares began a precipitous drop. Over the next 15 minutes, before Nasdaq halted trading, they dropped as low as $3.

And every bit as important:

The damage was exacerbated by the growing use on Wall Street of automated programs that trigger stock trades without any human interaction. The so-called algorithmic trading mechanisms, which buy and sell stocks based on news headlines and earnings data, were responsible for roughly a quarter of New York Stock Exchange trades in the last week of August.

Investors said simple human scrutiny would have indicated the UAL story was old, but computerized trading systems don't make such determinations.

I'm not really clear about what the authors are describing here -- is it just algos that respond to market prices or conditions, or is it the newer phenomenon of software that electronically "reads" wire stories and spits out orders in response? In fact, a number of news organizations sell newsfeeds custom-made to be read by such software. I wonder what role, if any, such readers played in the precipitous fall in UAL's stock price, and what safeguards the wires, traders, or software makers might have in place to prevent the software from being duped by hoaxes or erroneous six-year-old "news."

So let me see if I have this all straight:

1) A piece of software mistakes old news for something new, and the mistake is passed around electronically from site to site without the benefit of any human editor anywhere in the process to say, "Hey, wait a minute..."

2) The erroneous "news" is read and traded on by, at least in part, machines without the benefit of any human trader to say, "Hey, wait a minute..."

3) Those trades are sent to the stock market that actually prides itself on having no specialist at the switch to say, "Hey, wait a minute..."

4) We humans are shocked, shocked that such a thing has occurred.

It is no accident it occurred; obviously, we have designed a world in which the only surprise should be that such things don't happen more often.

And as I mentioned yesterday, we have a different model here: fast, but with people at the switch. We demonstrated that a little more than a year ago, when people here jumped in to prevent a big, erroneous order from getting very far, as discussed here at the time.