Closer Look at Composite Sentiment Score

With the development of an extensive list of categories covering more than 330 market moving events, RavenPack has produced a similar but independent Event Sentiment Score (ESS).

To verify additional diversification effects, we take a closer look at the event distribution of CSS and the cross-distribution between the CSS and ESS scores.

Intuitive relationship between CSS and events

In terms of sentiment direction, the CSS score is consistent with the actual events detected from the underlying news stories. For example, most of the "very negative" records according to CSS fall into the "Earnings-Negative" category.

Strong positive relationship between CSS and ESS.

There exists a strong positive relationship between CSS and ESS. Specifically, the correlation between the two analytics is 32.6%, while the Spearman's rank correlation is 43.7%. Both are significant at 0.1% confidence level.

ESS offers additional guidance beyond CSS.

When available, ESS offers additional guidance beyond CSS in classifying the sentiment direction of a news record. As an example, when an event is classified as neutral by CSS (CSS=50), with ESS there is a 76.2% chance that such event can be further classified as positive or negative (positive if ESS>50 or negative if ESS<50).

Introduction

This is the first edition of the new RavenPack Research Series. As part of our innovation strategy, we strive to deliver unique insights into the RavenPack datasets, and to provide inspiration on how news analytics can be applied in a quantitative investing and trading. In this report, we take a closer look at the RavenPack Composite Sentiment Score (CSS) and its relationship to the underlying news events measured by the RavenPack Event Sentiment Score (ESS).

CSS is considered a story-level sentiment analytic that represents the overall news sentiment by combining various sentiment analysis techniques, namely scores identified as PEQ, BEE, BMQ, BCA and BAM (see Appendix A for definitions). CSS can take values between 0 and 100, and can be linked to companies mentioned in the news via the RavenPack Relevance score (see Appendix A). Typically, CSS is applicable when a company’s Relevance in a news story is greater than 90, which is considered significant. In this study we focus on Relevance equal to 100, which is considered very significant as this means a company plays a key role in the news story.

Generally, the direction of CSS is determined by examining emotionally charged words and phrases and by matching stories typically rated by experts as having short-term positive or negative share price effect. The magnitude of the score is empirically determined by the size of the market impact of such event tested on intraday tick data from approximately 100 large-cap stocks. A score greater than 50 represents positive sentiment, a score below 50 negative sentiment, while 50 represents neutral strength...