“Machine readable news” services, new at Thomson Reuters

From the FT — some giant leaps for robot-kind in the world of trading:

The arms race in trading technology is set to intensify this week as Thomson Reuters, the news and market data company, on Monday unveils a service for “high-frequency” traders allowing them to make split-second trading decisions based on news articles “before the information moves the market” . . .

So-called “machine readable news” services, such as the new Thomson Reuters product, have grown up in parallel with the emergence of high-frequency and algorithmic trading, which depend on lightning-fast delivery of data and news to traders specialising in such computer-driven trading strategies.

Machine readable news systems use computers to “scrub” thousands of breaking news stories, prioritising their relevance for traders – often based on simple key words – and delivering them in a special feed. This provides traders with “signals” that are used to drive their strategies.

This is fascinating stuff — but it’s also something which throws open plenty of theoretical questions.

For instance, when Thomson Reuters-rival Bloomberg was first exploring the possibility of machine-readable news, it was rumoured be developing a system that would crudely code company stories with `buy’ or `sell’ signals. The problem, supposedly, was the labelling, or code-tagging of reports, by journalists. If a company released full-year results which beat expectations, but gave a negative outlook for the next financial year, would the code be `buy’ or `sell’ ?

The scope for human error on this type of news-coding is also massive. In the words of Bloomberg’s News-commandant, Matt Winkler, “the enemy is the human” here, and one that has the potential to create plenty of errors in this type of news-coding. Witness, for instance, the havoc wrought when the newswire republished an old story on United Airlines’ bankruptcy, sending shares crashing.

There is also the issue of whether trading bots will even find such data of use.

The new Thomson Reuters service is focused on futures and FX trading, where there’s been the suggestion that FX computer traders actually retreat (albeit briefly) from the market after an important news-release — non-farm payrolls, par example. The thinking is that many algorithms aren’t designed to deal with the sharp moves in exchange rates that take place right after news releases, even if they do have high-frequency capabilities.

Interestingly that hesitancy is something which news-reading systems can capitalise on. Thomson Reuters’ NewsScope Analytics checks out articles to determine whether they’re `relevant’ or `substantive’. If something big comes out on a company, computer traders can set up a sort of circuit-breaker which trips and pauses the computer trading programme.

What’s more, Thomson Reuters appear to have sidestepped the `human’ problem in the coding of news stories with buy or sell signals — using linguistic analysis to evaluate the `tone’ of the article

From the company’s Rich Brown, global business manager for machine-readable news:

Thomson Reuters NewsScope Analytics’ relevance indicator can be used to determine the article’s degree of “substantiveness” for a particular company. Feature or substantive articles on IBM [for example] would have a high relevance score while articles where IBM “shares” the story with a number of other companies would have a lower relevance score. The algo rule might suggest one turns the process back on when the news item has a relevance score below a specified level.

NewsScope Analytics also measures the sentiment, or tone, of the article. A very positively or very negatively-toned article might suggest a bias in the price to the upside-or downside, or to help predict a spike in volatility, which can then be incorporated into your strategies.

Furthermore, the novelty indicator in NewsScope Analytics takes a “vocabulary fingerprint” of the article and compares it to other articles on that company over various time periods. To the extent that the fingerprints look similar, the system notes how many times it has “heard” it before. If the item appears repetitious, the application logic might then trigger the algorithm to resume.

Anyway, this new Thomson-Reuters service is called NewsScope Direct, and it’s more about the latency, that is, the speed at which it functions, than the news-reading per se. Hence the targeting of high-frequency traders, not just algorithmic.

Here’s the press release:

Housed in Thomson Reuters London and Chicago hosting centers, NewsScope Direct allows clients to easily incorporate news and events into a variety of trading strategies. Customers can connect to the newsfeed from their own data centers or leverage Thomson Reuters proximity hosting solution. This will benefit high frequency trading firms operating in the FX and Futures markets which can move sharply in response to key news and economic data.

Housed in Thomson Reuters London and Chicago hosting centers, NewsScope Direct allows clients to easily incorporate news and events into a variety of trading strategies. Customers can connect to the newsfeed from their own data centers or leverage Thomson Reuters proximity hosting solution.This will benefit high frequency trading firms operating in the FX and Futures markets which can move sharply in response to key news and economic data.

Faster latency and news-reading capabilities?

Call us sufficiently scared impressed.



http://ftalphaville.ft.com/blog/2010/01/26/134561/rise-of-the-news-reading-machines/

 

Good overview in Traders mag - http://www.tradersmagazine.com/news/newsfeeds-investors-107666-1.html