The Next Generation of Sentiment Data

  • Thursday, September 14, 2017

Sentiment data has a nasty reputation for overpromising and under delivering. However, the work of Prattle and other emerging text analysis firms has put those days behind us.   

While the idea of knowing (and trading) on the quantified emotions of market actors before they have made their way into asset pricing is the dream of sentiment data, the technological reality has often fallen short. Over the last few years, sentiment analysis systems have used Twitter data as the standard for the “emotions” of the market and simplistic, fixed dictionaries of “good” and “bad” words as the foundation of their analysis. Unsurprisingly, the data produced by this approach were not clean, coherent, or remotely tradable. Consequently, “sentiment analytics” earned a bad reputation as “useless data.”

New Call-to-action

Unlike traditional sentiment analysis companies that sought to impose order on the chaos of tweets or news articles, Prattle approached the problem a very different way. In particular, Prattle focused on market-moving communications, like those from central banks. These complex, nuanced communications were previously deemed too challenging to evaluate with the dictionary-based approach. By starting with the most challenging language, “Fed-speak,” Prattle had no choice but to solve the problem in a wholly unique way.

Instead of standard dictionaries, Prattle algorithmically generates a lexicon tailored to each institution it evaluates. Instead of fixed dictionaries, Prattle uses machine learning to constantly update the lexicons. Instead of being populated by simple “good” and “bad” words, these algorithmically generated lexicons are composed of words, phrases, sentences—and sometimes whole paragraphs—that are assessed values relative to their actual, historical relationship to price. Prattle uses these lexicons as the foundation of its analysis of central bank communications, and the analytics it produces are clean, coherent, and directly tradable. Because it was developed for the most complex financially relevant language, this approach readily transfers to other market-moving communications, namely, corporate communications.

Prattle’s unique approach and the resulting analytics are getting recognized by established industry voices. Quinlan & Associates chose Prattle as the case study in a recent report on alternative data, and the San Francisco Fed has featured Prattle’s analytics in several research notes. As more respected institutions embrace this technology, a clear signal emerges: the next generation of sentiment data has arrived. For financial professionals, this is great news. Sentiment data has always held great potential, and Prattle is part of a new class of firms ensuring that the technology is ready to deliver.

For more on Prattle’s analytics and the underlying methodology, download our analytics Primers.

The Prattle Team