George presents his latest research on news sentiment signals in the framework of the MSCI Barra equity factor models. He evaluates factors constructed from the latest generation of RavenPack data, whereby highlighting improvements over previous versions in coverage, cross-sectional explanatory power and factor returns. George also demonstrates how the results are robust to factor formulation, geography, and time period.
Areas for Consideration
- Should We Upgrade From RPNA 3.X To RPA 1.0?
- Signal Construction
- From CAPM to Multi-Factor Models
- Correlation of News Sentiment with Other Style Factors is Low
- Coverage is Improved
- Performance of Basic 1M Average Sentiment with RPA 1.0 vs RPNA 3.X
- Relevance and Novelty Filters Add Value
- Coverage, Media News Counts Per Stock, ESS
- Decile Portfolios, Impact of Filters
In conclusion MSCI did find significant improvements with the new RPA 1.0 data set over our existing RPNA 3.X in coverage. Also it looks like their enhancements to the event detection algorithm and taxonomy really did improve the performance of the Event Sentiment Score drive signals and the addition of many more news sources drove the improved performance of the Composite Sentiment Score type of signals.
In general the additional filtering on relevance and novelty did improve the performance of signals in general and these results were quite consistent in the regions that we explored.
Access the full video and slides
Request access to the full video and slides of the session "Multi-Dimensional Analysis of News Sentiment Factors" held at the RavenPack Research Symposium in New York City in September 2018.Request Event Materials