Refining Twitter Trading Signals with Bayesian Probability

In March, we introduced a statistic to enhance our StockTwits S-Score signal called the Bayesian Accuracy Rating (BAR). The positive results prompted us to extend this methodology to our Twitter S-Factor signal, the longest-running product in our offerings. For those seeking technical details on the Bayesian Accuracy Rating, we refer you to our earlier blog post. In short, the BAR offers a probability metric derived from the historical sentiment scores and subsequent returns over the previous rolling one-year period.

Expanding our scope to stocks priced over $5, we adjusted our BAR filter threshold from .50 to .55 to enhance signal accuracy. We then compared the same two strategies: one using the traditional S-Score > 2 criteria, and the other adding the BAR > .55 requirement. Both strategies were benchmarked against the SPY index. Signals were captured at 9:10 AM, 20 minutes before market opening, with subsequent open-to-close returns calculated. Daily re-bucketing and compounding over five years provided a comprehensive view of strategy efficacy.

Over this period, the standard S-Score > 2 strategy outperformed the SPY index by 8%, with an average of 184 daily stocks. Introducing the Bayesian filter reduced bucket size to around 62 stocks but boosted cumulative returns by over 50%.

Year-to-date analysis showed the S-Score > 2 strategy outperforming the SPY index, while the Bayesian-filtered S-Score cohort tripled returns compared to the non-filtered group.

In summary, our exploration into Bayesian Probability has yielded promising results, confirming its role in enhancing signal predictability. By refining methodologies and leveraging Context Analytics social sentiment data, we continue to uncover opportunities for stock market returns. For information on our data visit www.contextanalytics-ai.com or click the button below!