Advances in Machine Learning for Computational Finance
               International Workshop, July 20-21 ‘09 London, U.K.

                  http://web.me.com/davidrh/AMLCF09/Workshop.html


                              Abstract for Invited Presentation

              Identification of Actionable Financial News
                        Using Machine Learning
       David Leinweber, UC Berkeley Haas School of Business, Leinweber & Co.
                            Jacob Sisk, Leinweber & Co.
                        Richard W. Brown, Thomson Reuters
                          William Fang, Thomson Reuters


Many previous studies of the effects on news on the prices of stocks and other securities
have found that price changes often precede the release of a news story. Yet we know
from daily experience that this is not true for all stories, and that market reaction
following some news reports can be large and exploitable.

There are many characteristics of news stories that are potentially useful in distinguishing
“old news” from “predictive news”. These include the type and source of news, subject
and firm characteristics, natural language measurements on content, such as sentiment,
and the context of the story in relation to others, and in relation to quantitative market
information.

Can machine learning methods, supervised and unsupervised, be used to distinguish
stories that are predictive of future price and volatility from those that are not? This talk
presents results from research on this topic sponsored by Thomson Reuters, one of the
world’s largest financial information firms. It draws on an unusually long and deep
history of both news and prices on intraday time scales.


Author Contact Information:
David Leinweber <djl@haas.berkeley.edu>, (corresponding author)
Jacob Sisk <jacob.sisk@gmail.com>,
Richard W. Brown" <richard.w.brown@thomsonreuters.com>,
William Fang <william.fang@thomsonreuters.com>

Rev. April 16, 2009

Advances in Machine Learning for Computational Finance

  • 1.
    Advances in MachineLearning for Computational Finance International Workshop, July 20-21 ‘09 London, U.K. http://web.me.com/davidrh/AMLCF09/Workshop.html Abstract for Invited Presentation Identification of Actionable Financial News Using Machine Learning David Leinweber, UC Berkeley Haas School of Business, Leinweber & Co. Jacob Sisk, Leinweber & Co. Richard W. Brown, Thomson Reuters William Fang, Thomson Reuters Many previous studies of the effects on news on the prices of stocks and other securities have found that price changes often precede the release of a news story. Yet we know from daily experience that this is not true for all stories, and that market reaction following some news reports can be large and exploitable. There are many characteristics of news stories that are potentially useful in distinguishing “old news” from “predictive news”. These include the type and source of news, subject and firm characteristics, natural language measurements on content, such as sentiment, and the context of the story in relation to others, and in relation to quantitative market information. Can machine learning methods, supervised and unsupervised, be used to distinguish stories that are predictive of future price and volatility from those that are not? This talk presents results from research on this topic sponsored by Thomson Reuters, one of the world’s largest financial information firms. It draws on an unusually long and deep history of both news and prices on intraday time scales. Author Contact Information: David Leinweber <djl@haas.berkeley.edu>, (corresponding author) Jacob Sisk <jacob.sisk@gmail.com>, Richard W. Brown" <richard.w.brown@thomsonreuters.com>, William Fang <william.fang@thomsonreuters.com> Rev. April 16, 2009