CoChat
Sign Up
Cited by 117Stock Market Forecasting MethodsFinancial Markets and Investment StrategiesSentiment Analysis and Opinion Mining

Sentiment trading with large language models

Open access
Kirtac, Kemal, Germano, Guido·2024·Finance Research Letters
DOI 10.1016/j.frl.2024.105227
Open this paper in CoChat — free accountOpen PDF

Abstract

We analyze the performance of large language models (LLMs) including OPT, BERT, and FINBERT, alongside the traditional Loughran-McDonald dictionary, in sentiment analysis of 965,375 U.S. financial news articles from 2010 to 2023. Our findings reveal that OPT significantly outperforms others, accurately predicting stock market returns with a sentiment prediction accuracy of 74.4%. A long-short strategy based on OPT with 10 bps transactions costs yields a superior Sharpe ratio of 3.05. From August 2021 to July 2023, this strategy produces an impressive 355% gain, outperforming other strategies and traditional market portfolios. This underscores the potential of LLMs to transform financial market prediction and portfolio management, and the necessity of employing sophisticated language models to develop effective investment strategies based on news sentiment.

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1016/j.frl.2024.105227

Cites

32 of 33 references verified0 citation errors · 0 not checkable

Statistics

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • LLMs specifically fine-tuned with financial-related datasets (FinGPT, OPT, etc.) demonstrated further improvement by aligning LLMs with domain-specific knowledge.

    “Additionally, [18, 56] researched the performance of LLMs(FinGPT, OPT, etc.) that specifically fine-tuned with financial related dataset and demonstrated further improvement by aligning LLMs with domain-specific knowledge.”

    From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [18]

    The Methods (§2.2; page unavailable) say the authors fine-tuned OPT and BERT for financial analysis and included FinBERT, which was already tailored to financial text. But the Results (§3.1) rank OPT first, BERT second, and FinBERT third; §3.2 raises possible overfitting in specialized models. The paper therefore supports studying these models, not the claim that aligning them with domain-specific knowledge demonstrated further improvement. (LLM, conf=0.87)

    Experimental — accuracy not yet measured

  • In [18, 29], the long-short strategy has shown to outperform both long-only and short-only strategies in backtesting.

    “A similar approach is adopted in [18, 29], where the long-short strategy has shown to outperform both long-only and short-only strategies in backtesting.”

    From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [18]

  • In both [18] and [29], portfolios weighted by market capitalization have shown slightly higher returns than those that are equally weighted.

    “In both[18] and [29], portfolios weighted by market capitalization have shown slightly higher returns than those that are equally weighted.”

    From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [18]

  • In [18] and [60], F1 score and accuracy are used to measure model's prediction accuracy of news sentiment.

    “In [18] and [60], the F1 score and accuracy are used to measure the model's prediction accuracy of news sentiment.”

    From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [18]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. X. · DOI · Paper page

Cited by

Sorted by citations

Related

Authors
Kirtac, K., & Germano, G.
Affiliations
  • University College London (GB)
  • Systemic Risk Centre (GB)
Corresponding author
Kirtac, Kemal
ORCID
Kirtac, Kemal
Venue
Finance Research Letters
Publisher
Elsevier BV
Published
2024-04
Volume
62
Pages
105227
ISSN
1544-6123 (ISSN-L)
DOI
10.1016/j.frl.2024.105227
OpenAlex
W4392859372
Cited by
115
References
30
Citations by year
2026: 58 · 2025: 45 · 2024: 12
FWCI
34.98
Type
journal article
Language
en
Open access
Licence
Locations
  • Finance research letters (publishedVersion, cc-by)
  • SSRN Electronic Journal (acceptedVersion)
  • London School of Economics and Political Science Research Online (London School of Economics and Political Science) (acceptedVersion)
  • RePEc: Research Papers in Economics (submittedVersion)
  • arXiv (Cornell University) (submittedVersion, public-domain)
  • UCL Discovery (University College London) (submittedVersion)
  • London School of Economics and Political Science Theses Online (London School of Economics and Political Science) (submittedVersion)
  • arXiv (Cornell University) (public-domain)
Funders
Economic and Social Research Council (ES/K002309/1, ES/R009724/1, ES/Y010612/1)
Topics
  • Social Sciences › Decision Sciences › Management Science and Operations Research › Stock Market Forecasting Methods
  • Social Sciences › Economics, Econometrics and Finance › Finance › Financial Markets and Investment Strategies
  • Physical Sciences › Computer Science › Artificial Intelligence › Sentiment Analysis and Opinion Mining
Keywords
large language models, OPT, BERT, FinBERT, financial news sentiment, stock return prediction, long-short strategies, Sharpe ratio

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review