Stocksnips Version 3.0 Sentiment Data Set Highlights
* Enhanced algorithms for improved accuracy of attribution of news snippets to companies. This has resulted in lower number of false positives.
* Sentiment signal now incorporates a memory-based attenuation model to ensure that the impact of past and current news is weighted appropriately. We conducted a study to determine reliability and construct validity of our signal. We considered the complexity related to the discrete and bounded nature of news sentiment. The results show that there is a high signal to noise ratio with strong construct validity. Our analysis also indicates a much higher correlation to price movement versus pure social media driven sentiment signals
* Coverage for over 2000+ US equities for 3 years (2016 – 2018)
* Improved message weighting for SEC information (versus pure news-based information sources)
* Improved message attenuation model (weighting of messages over-time based on current message volume and historical message signal)
* Optimization across randomly selected 640 stock tickers – to ensure Sentiment Signal reliability and construct validity (i.e. stock price vs. sentiment correlation)
* Resulted in 50% improvement in sentiment correlation with stock price across 2,000+ tickers
* Improved reliability with 60% reduction in “random” sentiment variability – (without sacrificing construct validity)
* Improved sectoral and market level correlation
* Overall, new V3 StockSnips sentiment signal provides statistically significant (p < 0.01) increase in stock price correlations – using better message weighting and attenuation heuristics.