Enhancing Alpha Generation: The Power of Combining Social Media and News Sentiment

At Context Analytics, we are at the forefront of unstructured financial data, providing innovative tools to transform raw information into actionable insights. Two of our primary data sources are social media platforms and news articles, with a particular focus on Twitter, Stocktwits, and financial related News. By processing thousands of tweets, messages, posts, and articles daily, we leverage our patented natural language processing (NLP) technology to generate standardized metrics, such as the S-Factor feed for social media and article-level sentiment metrics for news.

In this blog, we explore how combining these three data sources—Twitter, Stocktwits, and News—can enhance trading signals. Specifically, we examine the sentiment for individual securities across all three platforms. For social media, we use our S-Score from the S-Factor feed, which quantifies the sentiment of conversations around a stock (for more information on the S-Score, read this blog). For news, we aggregate positive phrases associated with each security over a 24-hour period to create a "positivity ratio," expanding on previous research on the impact of negativity ratio.

To start, we demonstrate that these three sources offer unique and uncorrelated signals. By analyzing the S-Scores for Twitter and Stocktwits, taken at pre-market close, alongside the news positivity ratio, we observe a Pearson correlation below 0.2 across the datasets. The correlation between Twitter and Stocktwits is higher, but there is almost no correlation between these two and news sentiment. For example, a highly positive day on Twitter does not necessarily indicate a similarly positive sentiment in the news. These uncorrelated factors enable us to combine the information from each source to create a more robust signal.

We tested this by building eight different daily rebalanced trading portfolios. These portfolios were based on daily quintiles of the sentiment metrics mentioned above. The strategies included long positions for the top quintile (20%) of stocks by S-Score in each data source—Twitter, Stocktwits, and news—and an additional long strategy that selected securities appearing in the top quintile across all three sources. Similarly, we created short portfolios based on the bottom quintiles of sentiment for each data source. By comparing these portfolios, we aimed to determine whether combining signals from all three sources enhances the overall trading performance by providing a sentiment consensus.

Our strategy involved trading U.S. equities with a price greater than $5, compounding returns daily. We began the analysis in August 2021, which is the earliest point in our news dataset, and continued through to the present.

Over this period, we found that using all three sources strengthened the signal for both long and short strategies. In the long portfolios, all top quintiles followed a similar positive return pattern, with the news-based strategy outperforming those of Twitter and Stocktwits. On the short side, all three sources produced similar negative returns.

What stood out most was the performance of the 12 to 15 stocks that ranked in the top quintile for all three sources—these "consensus positive" stocks significantly outperformed each of the individual top quintile portfolios while also increasing risk-adjusted returns. A similar outperformance was observed in the short portfolios. The unique perspectives provided by the three uncorrelated data sources contributed to a broader and more powerful signal, resulting in enhanced performance.

This analysis shows that while each individual source offers value, combining sentiment from multiple sources can amplify signals and improve trading strategies. At Context Analytics, our tools harness the power of unstructured data to generate alpha, and integrating multiple data sources helps traders gain a more comprehensive understanding of market sentiment.

For more information, visit www.contextanalytics-ai.com