How To Predict The News Before It Happens

Matt Rosoff | Jul. 22, 2011, 6:05 PM http://www.businessinsider.com/how-to-predict-the-news-before-it-happens-2011-7

Recorded Future takes tens of thousands of inputs from public sources like SEC filings, blog posts, and tweets, and compiles them to give its clients a glimpse into future events.

This afternoon, CEO Christopher Ahlberg posted a few tantalizing details about what they look for.

Looking months or weeks out, these signals are interesting:

Covert behavior: Like layoffs followed by massive insider trading.

Deceit signals: Words like "frankly," honestly," "confidentially," and "as you know" are strong signals that somebody is about to lie.

Statements with future dates: Phrases like "Apple is to release the iPhone 5 on September 7." Not every one of these statements will be accurate, but the aggregate of lots of them might be.

Obscure signals: A forum post like "this drug is making me sick" may provide information about a clinical drug trial ahead of time.

Unusual behavior: For instance, when two companies that haven't been mentioned together are suddenly appearing next to each other in news stories, or when two companies are suddenly quiet for two weeks straight, that might mean a merger is imminent.

Days and hours out, the signals get more explicit:

Directional media bursts like statements about Google growing rapidly and suddenly.

Rumors like "we hear AAPL will show great iPad sales tonight."

Direct statements about the future like "the president will hold a press conference on a matter of national security."

And there's sometimes a (brief) opportunity to detect an event between the time it happens and when it's first reported, like an explosion of tweets on the same subject or a blogger posting about a strange explosion in his town.

The trick is not only in detecting the signals, but also deciding which ones are believable and which aren't. That's where deep statistical analysis comes in -- and where Recorded Future believes it has an edge.