mentioned in market commentary (i.e., ‘‘Microsoft Down in Late Day Trading’’). This
can become a source of frustration and trading inefficiency when certain stocks’ circuit
breakers are tripped too often and/or for the wrong reasons. The problem is exacerbated
further when one’s algo rules state that news on Intel, Dell, and Microsoft (those stocks
that may be highly correlated with IBM’s) should also stop the algorithm executing
orders for IBM, or that news on IBM’s customers or suppliers should stop the algorithm
as well. And the problem gets exponentially worse as you add more and more sources of
information to this logic. A market-maker using this approach to widen its bid–ask
spreads when news is published relating to one of its companies would find a very low
signal-to-noise ratio and might find itself perpetually out of the market. Unable to keep
pace with reviewing news-tripped exceptions, one may eventually turn off the news logic
and again be exposed to unforeseen risk.
So, how does one increase the signal-to-noise ratio, ensuring protection from unforeseen
exposures without an excessive number of halts or items to review?
There are two defensive approaches to this problem: (1) trip the circuit breaker only
when substantive, important, ‘‘new’’, credible, or highly emotive news is published on a
company, or (2) trip the circuit breaker on any news about a company and turn the
algorithm back on after programmatically determining that the news is duplicative, non-
substantive, less emotive, or unimportant. Either approach can be substantially better
than many of the processes firms use today, but the second option is ideal for traders
who are making markets or are very sensitive to order fulfillment at the millisecond level.
This offers the best protection by tripping algorithms immediately, but allows for a
much quicker re-entry to market-making or order execution (measured in milliseconds).
It can also significantly reduce the number of items that require a more thorough review
by human algorithmic trading supervisors.
Sounds logical, right? So how exactly can this be done?
By using the metadata offered in a machine-readable newsfeed, one can quite easily
implement the above strategies. In the example outlined, the circuit breaker would
initially be triggered by the presence of a news story on IBM, widening the bid–ask
spread or stopping the algorithm altogether. Based on a set of rules, the algorithm would
then be reversed, resumed, or perhaps accelerated in the presence of certain news types.
Thomson Reuters News Analytics offer a robust set of metadata, which can be used to
help address this situation. For example, the ‘‘relevance’’ indicator is used to determine
the extent of an article’s focus on a particular company. Feature articles on IBM would
have a high relevance score while articles in which IBM is one of many companies
included would typically have a lower relevance score. If the circuit breaker is tripped
automatically, the algo rule might suggest that one turns it back on when the news item
has a relevance score below a specified level.
The sentiment, or tone, of the article can also be a powerful signal. A very positively
or very negatively-toned article might push the item to the top of a review list so that
a trader can more quickly take advantage of a likely price movement. Stopping an
aggressive purchase in the face of a very negatively toned item may give you significant
protection from being blindsided by an event, while accelerating your purchase in the
308 Industry insights, technology, products, and service providers