Identify the strongest instrument for case-level sentiment–return associations

Identify which of VADER, Loughran–McDonald, FinBERT, Twitter-RoBERTa, and Claude Haiku most reliably measures the association between event-minus-baseline changes in message negativity and cumulative abnormal returns at the securities-class-action case level.

Background

The case-level analysis correlates the change in average negativity between an event window and a baseline window with cumulative abnormal returns for 134 cases. Claude Haiku produces the largest observed association, but the comparison is underpowered and sensitive to how case-level negativity is defined.

Alternative constructions elevate Loughran–McDonald and Twitter-RoBERTa to a similar range, while restricting the analysis to relevant messages reduces the sample to 50 cases and makes every instrument nonsignificant. The paper therefore does not establish which instrument is strongest for this case-level relationship.

References

The robust content of this table is that a case level contrast exists and is several times larger than any daily panel effect; the identity of the strongest instrument at this level is not established.

— Same Day, Same Story; One Day Ahead, a Different Signal: The Dual Validity of Financial Sentiment  (2609.11144 - Aravinthkakshan et al., 10 Sep 2026) in Section 6.3, “Divergence at the case level”