Experiments in Conditioning Risk Estimates with Quantified News

Since 1997, Northfield has used various forms of “contemporaneously observable” information to improve estimates of the future risk of securities and investment portfolios.  The first effort was to include changes in “option implied” stock volatility in our short horizon risk models in the late 1990s.   In 2007, we began to expand this concept using information sets outside the regular data inputs to our models to adjust risk forecasts to fit current market conditions.  A “near horizon” family of models was commercially introduced in 2009 and uses observable broad market information (e.g. VIX level, TED spread) to inform our models on how things now are different from the way they usually are.  We have conducted joint research projects with two groups of MIT graduate students to explore how quantified news text can used to further improve risk estimates of over short time horizon, as first suggested in diBartolomeo, Mitra and Mitra (Quantitative Finance, 2009).   A recent internal research project confirms the efficacy of quantified news (T > 9) as an input to next day volatility estimation over a sample of 1.7 million data points.  This presentation will review the research done to date on “conditional” risk modeling, describe the various forms of quantified news feeds that are now available to investment professionals, and provide empirical data from the various studies on the extent to which quantified news has been demonstrated to be a useful ingredient of security level risk assessment.

Leave a Reply

Your email address will not be published. Required fields are marked *