The DOI matches its Crossref record; 58 of the 75 references it cites were found in a registry or an index, and 5 are not checked yet; 45 of the 75 sources its claims cite were checked for support.
Large Language Model Agent in Financial Trading: A Survey
Open accessAbstract
Trading is a highly competitive task that requires a combination of strategy, knowledge, and psychological fortitude. With the recent success of large language models (LLMs), it is appealing to apply the emerging intelligence of LLM agents in this competitive arena and understanding if they can outperform professional traders. In this survey, we provide a comprehensive review of the current research on using LLMs as agents in financial trading. We summarize the common architecture used in the agent, the data inputs, and the performance of LLM trading agents in backtesting as well as the challenges presented in these research. This survey aims to provide insights into the current state of LLM-based financial trading agents and outline future research directions in this field.
Verified 2026-10-10 · 5 references not checked yet
Overview
CoChat Summary
TL;DR
This survey reviews 27 papers on large language model (LLM) agents for financial trading, organizing their architectures, data inputs, and backtesting evaluations. It finds strong reported backtest performance but highlights limitations in model openness, data coverage, evaluation scope, and integration with real trading systems.
Key findings
- LLM trading agents fall into two main architectures: LLM as a Trader, including news-driven, reflection-driven, debate-driven, and reinforcement-learning-driven agents, and LLM as an Alpha Miner, where the LLM generates alpha factors for downstream trading systems.
- All reviewed agents use textual financial data; data categories include numerical, textual, visual, and simulated data, but only SEP incorporates real-time social media data, and FinAgent is the only reviewed work extending backtesting to cryptocurrency, specifically ETH.
- Across surveyed works, LLM agents achieved annualized returns ranging from 15% to 30% over the strongest baseline during backtesting with real market data.
- Among 14 papers using real market data for backtesting, 9 focus on the US stock market and 5 on the Chinese market; the median backtesting period is only 1.3 years, and few studies consider trading costs.
Methods
- The authors conducted a survey of 27 papers studying LLMs for financial trading, identified through multiple Google Scholar searches using keywords such as “LLM for trading” and “GPT stock agent,” with each paper manually assessed for relevance.
- The review categorizes agent architectures, data inputs, evaluation metrics, backtesting settings, baselines, and performance, drawing on the surveyed papers’ reported results.
Limitations
- The paper states that most agents rely on closed-source models such as GPT-3.5/GPT-4, raising data privacy and customization concerns; most studies use in-context learning without fine-tuning, with only one surveyed work tuning the LLM during training.
- The paper states that inference latency may make these models impractical for high-frequency trading, integration with existing trading systems is rarely discussed, few agents use social media data, and backtesting is mostly confined to US and Chinese stock markets with short periods and limited consideration of trading costs.
- The text makes evident that the survey’s performance conclusions depend on the reviewed papers’ backtesting setups, which the authors describe as often short, single-period, and with arbitrarily chosen start and end dates.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
27 Valid · 5 Not valid · 16 Not checked · 34 Own results and interpretations
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Existing works [29, 50] evaluate both close-source LLMs(e.g GPT3.5/4) and open-source LLMs(e.g Qwen[3], Baichuan[52], etc.) in financial sentiment analysis.”
What the sources say
- Could not verifycites [29] [29] Alejandro Lopez-Lira and Yuehua Tang. 2023. Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models. arXiv:2304.07619 [q-fin.ST] https://arxiv.org/abs/2304.07619 - sourceExperimental — accuracy not yet measured
The Abstract and 5.2 ScienceQA sections (page numbers unavailable) discuss visual instruction tuning and science question answering, not financial sentiment analysis. Searches for financial sentiment analysis, Qwen, Baichuan, and comparisons of closed- and open-source LLMs found no passage addressing the attributed claim. (LLM, conf=0.78)
- Not matchedcites [50] [50] Ruoxu Wu. 2024. Portfolio Performance Based on LLM News Scores and Related Economical Analysis. (January 30 2024). https://ssrn.com/abstract=4709617 -
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites [3] [3] Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, Keming Lu, Jianxin Ma, Rui Men, Xingzhang Ren, Xuancheng Ren, Chuanqi Tan, Sinan Tan, Jianhong Tu, Peng Wang, Shijie Wang, Wei Wang, Shengguang Wu, Benfeng Xu, Jin Xu, An Yang, Hao Yang, Jian Yang, Shusheng Yang, Yang Yao, Bowen Yu, Hongyi Yuan, Zheng Yuan, Jianwei Zhang, Xingxuan Zhang, Yichang Zhang, Zhenru Zhang, Chang Zhou, Jingren Zhou, Xiaohuan Zhou, and Tianhang Zhu. 2023. Qwen Technica source
The citation to this report is attached specifically to “Qwen” as an example of an open-source LLM, not to the claim about financial sentiment analysis. The Introduction states that the authors open-sourced the Qwen models, supporting that identification. (LLM, conf=0.93)
- Supportedcites [52] [52] Aiyuan Yang, Bin Xiao, Bingning Wang, Borong Zhang, Ce Bian, Chao Yin, Chenxu Lv, Da Pan, Dian Wang, Dong Yan, Fan Yang, Fei Deng, Feng Wang, Feng Liu, Guangwei Ai, Guosheng Dong, Haizhou Zhao, Hang Xu, Haoze Sun, Hongda Zhang, Hui Liu, Jiaming Ji, Jian Xie, JunTao Dai, Kun Fang, Lei Su, Liang Song, Lifeng Liu, Liyun Ru, Luyao Ma, Mang Wang, Mickel Liu, MingAn Lin, Nuolan Nie, Peidong Guo, Ruiyang Sun, Tao Zhang, Tianpeng Li, Tianyu Li, Wei Cheng, Weipeng Chen, Xiangrong Zeng, Xiaochuan Wang, Xiaoxi Chen, Xin Men, Xin Yu, Xuehai Pan, Yanjun Shen, Yiding Wang, Yiyu Li, Youxin Jiang, Yuchen source
The citation to this paper is attached to “Baichuan” as an example of an open-source LLM, not to the claim about financial sentiment analysis as a whole. The Introduction (page unavailable) identifies Baichuan 2 as a series of language models and says the models are open-sourced. (LLM, conf=0.93)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“However, SEP [19] is the only work we've reviewed that incorporates real-time social media data in LLM trading agent.”
What the sources say
- Partly supportedcites [19] [19] Kelvin J.L. Koa, Yunshan Ma, Ritchie Ng, and Tat-Seng Chua. 2024. Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models. In Proceedings of the ACM on Web Conference 2024 (WWW '24) . ACM. https://doi.org/10.1145/3589334.3645611 - sourceExperimental — accuracy not yet measured
Section 3.1.2 (p. 3) says SEP uses tweets collected through the Twitter API as inputs for stock predictions. It describes a dataset from 2020–2022, not a real-time social-media feed. The source also cannot establish the citing authors’ comparison with every other work they reviewed. (LLM, conf=0.86)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“Reinforcement learning methods, such as RLHF [37] and RLAIF [22], have proven effective in aligning LLM outputs with expected behaviors.”
What the sources say
- Supportedcites [37] [37] Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. 2022. Training language models to follow instructions with human feedback. arXiv:2203.02155 [cs.CL] https://arxiv.org/abs/2203.02155 - source
The citation to this report is attached to RLHF, not RLAIF. The Model-Assisted Safety Pipeline (p. 12) says RLHF produces responses better aligned with user intent, and Model Mitigations (p. 62) says it makes the models significantly safer, while noting that undesired behavior remains. (LLM, conf=0.91)
- Supportedcites [22] [22] Harrison Lee, Samrat Phatale, Hassan Mansoor, Thomas Mesnard, Johan Ferret, Kellie Lu, Colton Bishop, Ethan Hall, Victor Carbune, Abhinav Rastogi, and Sushant Prakash. 2023. RLAIF: Scaling Reinforcement Learning from Human Feedback with AI Feedback. arXiv:2309.00267 [cs.CL] https://arxiv.org/abs/2309. 00267 - source
The Introduction describes RLHF as effective for aligning language models with human preferences. The Conclusion reports that RLAIF achieved comparable improvements to RLHF on three text-generation tasks and that humans preferred their outputs at similar rates, supporting the claim about RLAIF’s effectiveness. (LLM, conf=0.91)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“This framework has demonstrated superior performance in backtesting compared to other agents including FinMem.”
What the sources say
- Supportedcites [57] [57] Wentao Zhang, Lingxuan Zhao, Haochong Xia, Shuo Sun, Jiaze Sun, Molei Qin, Xinyi Li, Yuqing Zhao, Yilei Zhao, Xinyu Cai, Longtao Zheng, Xinrun Wang, and Bo An. 2024. A Multimodal Foundation Agent for Financial Trading: ToolAugmented, Diversified, and Generalist. arXiv:2402.18485 [q-fin.TR] https: //arxiv.org/abs/2402.18485 - source
Section 5.3 identifies FinMem as one of the agents evaluated against FinAgent. Section 6 (Experimental Results) reports that FinAgent significantly outperformed the baselines overall, while noting that FinMem achieved higher returns on the ETH cryptocurrency dataset. (LLM, conf=0.90)
boostedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Given that classification models can be used for news sentiment prediction, machine learning or deep learning models such as Random Forest[30], LightGBM[17], LSTM[13], and BERT[7] are also used as baselines.”
What the sources say
- Supportedcites [30] [30] Gilles Louppe. 2015. Understanding Random Forests: From Theory to Practice. arXiv:1407.7502 [stat.ML] https://arxiv.org/abs/1407.7502 - source
The citation is attached to “Random Forest,” not to the other named models or the citing authors’ choice of baselines. The paper’s random-forests discussion (p. 72) describes Random Forests as a machine-learning method, supporting that identification. (LLM, conf=0.92)
- Supportedcites [17] [17] Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. 2017. LightGBM: A Highly Efficient Gradient Boosting Decision Tree. In Advances in Neural Information Processing Systems , I. Guyon, U. Von Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vishwanathan, and R. Garnett (Eds.), Vol. 30. Curran Associates, Inc. https://proceedings.neurips.cc/paper\_ files/paper/2017/file/6449f44a102fde848669bdd9eb6b76fa-Paper.pdf - source
The citation is attached to LSTM, not to the other listed models. Section 5.1 (p. 13) identifies LSTM as one of the neural networks used in the paper; using it as a baseline for news-sentiment prediction is the citing authors’ choice, not a result attributed to Jiang et al. (LLM, conf=0.88)
- Could not verifycites [13] [13] Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long Short-Term Memory. Neural Comput. 9, 8 (nov 1997), 1735-1780. https://doi.org/10.1162/neco.1997.9. 8.1735 - sourceExperimental — accuracy not yet measured
The citation is attached to LSTM in the citing sentence. The source’s Experimental Setup (p. 6) describes GPT-2 and GPT-3 models for mathematical answer generation; searches for LSTM and long short-term memory found no passage addressing their use as news-sentiment baselines. (LLM, conf=0.76)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
29 more claims · 8 with their cited sources checked
Create a free account to see all 82 claims and to verify the 21 references still open.
- Probably not a substantial claim. 33
- It may state two separate claims as one. 19
- The quote only partly states this claim. 1
Citations
Cites
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805 · Cited by 73,173
Proximal Policy Optimization Algorithms
Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017) · arXiv · doi:10.48550/arxiv.1707.06347 · Cited by 10,625
Playing Atari with Deep Reinforcement Learning
Mnih, V., Kavukcuoglu, K., Silver, D., Graves, A., Antonoglou, I., Wierstra, D., & Riedmiller, M. (2013) · arXiv · doi:10.48550/arxiv.1312.5602 · Cited by 9,471
Training language models to follow instructions with human feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · arXiv · doi:10.48550/arxiv.2203.02155 · Cited by 4,208
Sharpe, W. F. (1966) · The Journal of Business · doi:10.1086/294846 · Cited by 3,937 · cited 1×
“Sharpe Ratio [44] and Maximum Drawdown are used to assess the risk of the trading performance.”
Training Language Models to Follow Instructions with Human Feedback
Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., et al. (2022) · Advances in Neural Information Processing Systems 35 · doi:10.52202/068431-2011 · Cited by 3,117
Generative Agents: Interactive Simulacra of Human Behavior
Park, J. S., O'Brien, J. C., Cai, C. J., Morris, M. R., Liang, P., & Bernstein, M. S. (2023) · arXiv · doi:10.48550/arxiv.2304.03442 · Cited by 2,173
A survey on large language model based autonomous agents
Wang, L., Ma, C., Feng, X., Zhang, Z., Yang, H., Zhang, J., Chen, Z., Tang, J., Chen, X., Lin, Y., et al. (2024) · Frontiers of Computer Science · doi:10.1007/s11704-024-40231-1 · Cited by 1,941
A Survey on Large Language Model based Autonomous Agents
Wang, L., Ma, C., Feng, X., Zhang, Z., Yang, H., Zhang, J., Chen, Z., Tang, J., Chen, X., Lin, Y., et al. (2023) · arXiv · doi:10.48550/arxiv.2308.11432 · Cited by 1,925
Cited by
Potential of large language models in blockchain-based supply chain finance
Yang, L., Hou, Q., Xin, Z., Lu, Y., & Xu, L. D. (2025) · Enterprise Information Systems · doi:10.1080/17517575.2025.2541199 · Cited by 55
LLM-Powered AI Agent Systems and Their Applications in Industry
Liang, G., & Tong, Q. (2025) · IEEE World AI IoT Congress (AIIoT) · doi:10.1109/aiiot65859.2025.11105299 · Cited by 69
“These agents can process text, images and audio, making them applicable across diverse industries, including healthcare [44], finance [45], and manufacturing [46]–[48].”
Empowered stock market forecasting using Large Language Model on social media content
Sperlí, G., & Sichinolfi, M. A. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.112727 · Cited by 9
GenAI and LLM for Financial Institutions: A Corporate Strategic SurveyPreprint
Xu, J. (2024) · SSRN Electronic Journal · doi:10.2139/ssrn.4988118 · Cited by 8
DeepSeek: implications for data science and management in the AI era
Xu, Z., Liu, S., Huang, W., Shi, J., Xu, F., Tang, X., & Lu, H. (2025) · Data Science and Management · doi:10.1016/j.dsm.2025.09.001 · Cited by 8
Metadata
- Authors
- Ding, H., Li, Y., Wang, J., Chen, H., Guo, D., & Zhang, Y.
- Affiliations
- Columbia University (US)
- New York University (US)
- Venue
- Proceedings of the 2026 9th International Conference on Computers in Management and Business
- Publisher
- ACM
- Published
- 2026-02-27
- Pages
- 76-83
- OpenAlex
- W4402426816
- Cited by
- 10
- References
- 60
- Citations by year
- 2026: 2
- FWCI
- 34.01
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Locations
- International Conference on Computers in Management and Business (ICCMB) (publishedVersion, cc-by)
- RePEc: Research Papers in Economics (submittedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- arXiv (Cornell University) (cc-by)
- 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
- Social Sciences › Business, Management and Accounting › Management Information Systems › FinTech, Crowdfunding, Digital Finance
- Keywords
- large language models, LLM agents, financial trading, trading agents, backtesting, trading strategies, agent architecture
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.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
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
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Large Language Model Agent in Financial Trading: A Survey. https://app.cochat.ai/papers/10.1145/3802463.3802475
- The paper
- Ding, H., Li, Y., Wang, J., Chen, H., Guo, D., & Zhang, Y. (2026). Large Language Model Agent in Financial Trading: A Survey. Proceedings of the 2026 9th International Conference on Computers in Management and Business. https://doi.org/10.1145/3802463.3802475
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.