Financial NLP
Financial PhraseBank
Three-class sentiment on financial news sentences annotated by finance readers.
As of 13 August 2026. Conditional pack. Malo et al. 2014. Three-class financial-news sentiment. Recheck terms before commercial use.
Financial PhraseBank is 4,845 news sentences labeled positive, negative, or neutral by 16 finance readers. Accuracy and macro F1 are different when classes are unbalanced. A monitoring product that claims tone needs this diagnostic and a live labeled stream, not this pack alone.
No official Fable 5 or GPT-5.6 table is cited. The pack is in the catalog because the market still uses it, not because it is a 2026 frontier ladder.
- Scale
- 4,845 sentences
- Provenance
- Malo et al., 2014. License recheck required.
- Contamination
- High. Public since 2014. The most copied financial-sentiment baseline.
- Cannot establish
- A research-product alpha, monitoring quality, or economic value. Recheck Malo et al. terms before commercial use.
Metrics, kept separate
- Classification accuracy
- Macro F1
- Per-class F1
No cited public figure is installed for this pack. Method, scale, and gates still apply. A missing chart is not a pass rate.
What the figures bound
The original annotator agreement is the human ceiling for this three-class task. Later LoRA papers claim small models can match GPT-4 on this distribution. Those claims stay in their papers until replayed on a frozen revision.
Sentiment on news sentences is not filing QA, not alpha, and not a research-terminal score.
What this page does not conclude
- A 2026 frontier sentiment ranking
- Trading-signal or alpha quality
- A commercial run before terms are approved
- GPT-4 or FinGPT superiority without the cited paper's exact split
Cited as calibration on these memos — not a score of those products.
Not attached to a software, advisory, or automation memo yet.