21 October 2017

44975

Retired Member

3,171Posts 11,354,328Views 3,408Comments

Quantifying the News

24 February 2010  |  2426 views  |  0

Real time news technology has come a long way since the carrier pigeon.

Bloomberg just announced it is stepping up its presence in the hot machine-readable news scene. Dow Jones is there and Reuters is too, having recently announced its expansion of NewsScope Direct. These prolific financial news producers are capturing a growing trend among the algos and the high-frequency trading set: get the news a few microseconds before everyone else and let the black box figure out what to do with it.

It’s rather quaint to see content as king in this neck of the woods.

High-speed machine-readable news production is more about brute force computing than esoteric science. Extract news data elements in real time and distribute them really fast as news hits the wire.

The real action is in the black box. Or so it seems. In the span of a tiny fraction of a second, the black box interprets news data and predicts market impact before making the trade. Neat trick. How do they do it? At this point, news interpretation is straight forward. The machine-readable data points are known market-moving events, such as a macro indicator or a credit downgrade. The quants comb through archived news and historical market data, looking for patterns in past interactions between event announcements and market moves. Then they program the black box to take action in the real-time market state of play when a trigger event hits the wire.

The black box is not quite interpreting meaning algorithmically. It can’t yet rationalize the nuance of the written word. Though long the holy grail of content processing, algorithmic sentiment remains elusive, just out of grasp.

(Tagging an earnings announcement that has the words “earnings beat estimates” as positive doesn’t count!)

Many are trying are trying to get there, and a few are getting close. Companies such as RavenPackInfonicPsydex and even complex event processor Streambase are in the market with semantic technologies that interpret real-time unstructured content streams, such as Twitter, and produce machine-readable metadata about the content. But it remains awfully difficult to get right with enough certainty to make trades or big decisions with confidence.

Tags

Comments: (0)

Comment on this story (membership required)

Retired's profile

job title
location
member since 2014
Summary profile See full profile »

Retired's expertise

Member since 2009
3119 posts3,408 comments
What Retired reads

Who's commenting on Retired's posts

Ketharaman Swaminathan
Dharmesh Mistry
Nicola Cowburn
Michael Wright
Charmaine Oak
Francis Chlarie
Raymond Lee
Deepthi Rajan
Melvin Haskins
João Bohner
Bob Lyddon