Source: Huang, Wang & Yang (2023) — “FinBERT: A Large Language Model for Extracting Information from Financial Text”
I like finance papers that accidentally become papers about human behaviour. This one does that. It looks like a study about data, markets, or disclosure, but underneath it is really about how people decide what to believe.
One of the persistent challenges in applying NLP to finance is that financial language doesn’t work the same way as general language. Words that are negative in everyday use — “liability,” “risk,” “uncertainty” — often appear in neutral or even positive contexts in financial documents. Models trained on general text import this misalignment and measure sentiment incorrectly as a result. FinBERT is a domain-adapted version of BERT specifically trained on financial text. The researchers show that it substantially outperforms both traditional dictionary-based approaches like the Harvard Psychosociological Dictionary and general-purpose machine learning models — including Naive Bayes, SVM, Random Forest, CNN, and LSTM — on financial sentiment classification. The performance gap is most pronounced in two situations. First, when the sentence contains financial words not commonly used in general text: specialised terminology that general models haven’t learned to read correctly. Second, when the training sample is small: FinBERT’s pre-training on financial text allows it to generalise better from limited labelled examples, which matters for research settings where hand-labelling is expensive. The paper also shows that other models underestimate the textual informativeness of earnings conference calls by at least 18% compared to FinBERT.
In plain English, that is why the result matters beyond the chart. It changes where people should look, what they should question, and which comfortable assumption probably needs to be retired.
My takeaway: AI and data do not remove judgment from finance. They move judgment to the part where people decide what the signal means, whether it is trustworthy, and whether they are brave enough to act on it.