A Multimodal Foundation Agent for Financial Trading: Tool-Augmented, Diversified, and Generalist
Open accessAbstract
Financial trading is a crucial component of the markets, informed by a multimodal information landscape encompassing news, prices, and Kline charts, and encompasses diverse tasks such as quantitative trading and high-frequency trading with various assets. While advanced AI techniques like deep learning and reinforcement learning are extensively utilized in finance, their application in financial trading tasks often faces challenges due to inadequate handling of multimodal data and limited generalizability across various tasks. To address these challenges, we present FinAgent, a multimodal foundational agent with tool augmentation for financial trading. FinAgent's market intelligence module processes a diverse range of data-numerical, textual, and visual-to accurately analyze the financial market. Its unique dual-level reflection module not only enables rapid adaptation to market dynamics but also incorporates a diversified memory retrieval system, enhancing the agent's ability to learn from historical data and improve decision-making processes. The agent's emphasis on reasoning for actions fosters trust in its financial decisions. Moreover, FinAgent integrates established trading strategies and expert insights, ensuring that its trading approaches are both data-driven and rooted in sound financial principles. With comprehensive experiments on 6 financial datasets, including stocks and Crypto, FinAgent significantly outperforms 12 state-of-the-art baselines in terms of 6 financial metrics with over 36% average improvement on profit. Specifically, a 92.27% return (a 84.39% relative improvement) is achieved on one dataset. Notably, FinAgent is the first advanced multimodal foundation agent designed for financial trading tasks.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
3 of 7 cited sources checked · 1 supported · 0 flagged · accuracy not yet measured
- Premisecitation
The full capabilities of LLMs as comprehensive autonomous trading systems remain underexplored, particularly in their ability to interpret multimodal data and utilize diverse tools.
“Although FinMEM [56] is an LLM trading agent with a humanaligned memory mechanism and character design, the full capabilities of LLMs as comprehensive autonomous trading systems remain underexplored, particularly in their ability to interpret multimodal data and utilize diverse tools.”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [56] 56. Yangyang Yu, Haohang Li, Zhi Chen, Yuechen Jiang, Yang Li, Denghui Zhang, Rong Liu, Jordan W. Suchow, and Khaldoun Khashanah. 2023. FinMem: A performance-enhanced LLM trading agent with layered memory and character design. arXiv:2311.13743 [q-fin.CP] source
The Introduction and §2.2 discuss multimodal LLM agents, visual interpretation, and tool use in smartphone and other non-trading tasks. Our copy does not address whether comprehensive autonomous trading systems remain underexplored; searches for trading, financial markets, and that specific research gap found no relevant passage. (LLM, conf=0.76)
Experimental — accuracy not yet measured
- Findingown data
FinAgent achieves a 92.27% return (a 84.39% relative improvement) on one dataset.
“Specifically, a 92.27% return (a 84.39% relative improvement) is achieved on one dataset.”
- Also stated in: § Data · plain
“Specifically, a 92.27% return (a 84.39% relative improvement) is achieved on one dataset.”
Quote in the paperNot a substantial claim - Also stated in: § Data · plain
- Findingown data
The improvement from high-level reflection comes with a 7% rise in MDD% for TSLA.
“This improvement comes with a minor trade-off: a slight 7% rise in MDD% for TSLA.”
Quote in the paperNot a substantial claim - Findingown data
Across all datasets, FinAgent is the only method that consistently outperforms the broader market in terms of profitability.
“Across all datasets, FinAgent is the only method that consistently outperforms the broader market in terms of profitability.”
Quote in the paperSubstantial claim - Findingown data
FinAgent's performance surpasses other methods regarding cumulative returns, particularly on the TSLA dataset.
“Figure 4 illustrates that FinAgent's performance surpasses other methods regarding cumulative returns, particularly on the TSLA dataset.”
Quote in the paperSubstantial claim
- Findingreasoning
FinAgent yields slightly lower returns than FinMem on the cryptocurrency ETH.
“It's important to note that our approach yields slightly lower returns than FinMem on the cryptocurrency ETH, primarily because our auxiliary agents are specialized strategies tailored for stocks, not for cryptocurrencies with higher trading frequency.”
Quote in the paperNot a substantial claim - Findingown data
Adding augmented tools to MLH gives minor improvement in stock profitability.
“Compared to MLH and MLHT, there's a minor improvement in stock profitability.”
Quote in the paperNot a substantial claim - Findingreasoning
FinAgent anticipated a significant stock price drop post-September 14, 2023.
“Leveraging market intelligence and the reflection mechanism, FinAgent anticipates a significant stock price drop post-September 14, 2023.”
Quote in the paperNot a substantial claim - Findingown data
ETH performance dropped by over 20% due to the introduction of rule-based auxiliary agents specialized for stocks.
“However, the performance of ETH cryptocurrency dropped by over 20% due to the introduction of rule-based methods as auxiliary agents, which are specialized only for stocks.”
Quote in the paperSubstantial claim - Findingown data
On AAPL, diversified retrieval improves ARR and SR.
“As shown in Figure 5(a), we compare the performance of FinAgent with or without diversified retrieval on AAPL, and find that the use of diversified retrieval can contribute an obvious improvement in ARR and SR.”
Quote in the paperSubstantial claim - Interpretationreasoning
The lower returns on ETH are primarily because auxiliary agents are specialized for stocks, not cryptocurrencies with higher trading frequency.
“It's important to note that our approach yields slightly lower returns than FinMem on the cryptocurrency ETH, primarily because our auxiliary agents are specialized strategies tailored for stocks, not for cryptocurrencies with higher trading frequency.”
Quote in the paperSubstantial claim- Rests on an assumption: Cryptocurrencies have higher trading frequency than stocks.
- Interpretationown data
Stock-specific rule-based methods are ineffective for cryptocurrencies.
“As shown in Table 4 and Table 5, the 16% ARR% for solely T method starkly contrasts with the 29% ARR% of B&H in ETHUSD, highlighting the inefficacy of the stock-specific rule-based methods for cryptocurrencies and demonstrating that introducing to FinAgent significantly affects performance.”
Quote in the paperSubstantial claim - Interpretationnone given
FinAgent's market intelligence module extracts key insights from multimodal datasets encompassing asset prices, visual representations, news, and expert analyses, offering a multifaceted view across various markets.
“We introduce the market intelligence module, which is able to extract key insights from multimodal datasets encompassing asset prices, visual representations, news, and expert analyses, offering a multifaceted view across various markets.”
Quote in the paperSubstantial claim- Rests on an assumption: Insufficient multimodal data processing ability is a challenge for financial trading.
- Interpretationreasoning
FinAgent optimizes performance by balancing risk and reward effectively.
“It allows FinAgent to optimize performance by balancing risk and reward effectively.”
Quote in the paperSubstantial claim- Rests on an assumption: High returns often come with high risks.
- Interpretationnone given
FinAgent's decision-making module incorporates expert knowledge (supplementary expert guidance and auxiliary expert strategies) to guide decisions.
“Finally, the decision-making module incorporates expert knowledge, comprising both supplementary expert guidance and auxiliary expert strategies, to guide the agent's decisions.”
Quote in the paperSubstantial claim- Rests on an assumption: Integration of domain knowledge is a challenge for financial trading.
- Premisecitation
GPT-4 shows improved performance over GPT-3.5.
“As demonstrated by FinMem [56], OpenAI's GPT-4 shows improved performance over GPT-3.5.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites FinMem [56]
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Interpretationnone given
FinAgent employs a suite of tools including expert guidance and technical indicator-based advanced trading strategies to infuse domain knowledge in financial trading.
“We employ a suite of tools in FinAgent, including expert guidance and technical indicator-based advanced trading strategies, to infuse domain knowledge in financial trading.”
Quote in the paperSubstantial claim- Rests on an assumption: Integration of domain knowledge is a challenge for financial trading.
- Interpretationreasoning
FinAgent can take on slightly higher risk to achieve substantially greater returns.
“Therefore, FinAgent can take on slightly higher risk to achieve substantially greater returns.”
Quote in the paperSubstantial claim- Rests on an assumption: High returns often come with high risks.
- Interpretationown data
Introducing augmented tools significantly affects FinAgent's performance.
“As shown in Table 4 and Table 5, the 16% ARR% for solely T method starkly contrasts with the 29% ARR% of B&H in ETHUSD, highlighting the inefficacy of the stock-specific rule-based methods for cryptocurrencies and demonstrating that introducing to FinAgent significantly affects performance.”
Quote in the paperSubstantial claim - Interpretationreasoning
FinAgent represents a slight compromise on risk control.
“Hence, FinAgent represents a slight compromise on risk control.”
Quote in the paperNot a substantial claim- Rests on an assumption: High returns often come with high risks.
- Classified as probably not a substantial claim. 27
- May combine two claims. 24
Probably not a substantial claim: may not pass S3 (load-bearing).
PremisecitationLLMs have demonstrated remarkable capabilities in analyzing and interpreting financial data, as evidenced by developments like BloombergGPT and FinGPT.
“LLMs have demonstrated remarkable capabilities in analyzing and interpreting financial data, as evidenced by developments like BloombergGPT [48], and FinGPT [50].”
Quote in the paperNot a substantial claim1 of 2 sources checked- cites [48] 48. Shijie Wu, Ozan Irsoy, Steven Lu, Vadim Dabravolski, Mark Dredze, Sebas- tian Gehrmann, Prabhanjan Kambadur, David Rosenberg, and Gideon Mann. 2023. Bloomberggpt: A large language model for finance. arXiv preprint arXiv:2303.17564 (2023).
- cites [50] 50. Hongyang Yang, Xiao-Yang Liu, and Christina Dan Wang. 2023. FinGPT: Open- Source Financial Large Language Models. arXiv preprint arXiv:2306.06031 (2023). source
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
PremisecitationFinMEM is an LLM trading agent with a human-aligned memory mechanism and character design.
“Although FinMEM [56] is an LLM trading agent with a humanaligned memory mechanism and character design, the full capabilities of LLMs as comprehensive autonomous trading systems remain underexplored, particularly in their ability to interpret multimodal data and utilize diverse tools.”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [56] 56. Yangyang Yu, Haohang Li, Zhi Chen, Yuechen Jiang, Yang Li, Denghui Zhang, Rong Liu, Jordan W. Suchow, and Khaldoun Khashanah. 2023. FinMem: A performance-enhanced LLM trading agent with layered memory and character design. arXiv:2311.13743 [q-fin.CP] source
Section 2.2 discusses LLM agents generally, while Section 3.2 describes an agent that operates smartphone apps. Searches for FinMEM, trading, and the claimed memory mechanism and character design found no passage addressing the claim in our copy. (LLM, conf=0.76)
Experimental — accuracy not yet measured
May combine two claims.
Findingown dataFinAgent significantly outperforms 12 state-of-the-art baselines on 6 financial datasets (including stocks and Crypto) across 6 financial metrics, with over 36% average improvement on profit.
“With comprehensive experiments on 6 financial datasets, including stocks and Crypto, FinAgent significantly outperforms 12 state-of-the-art baselines in terms of 6 financial metrics with over 36% average improvement on profit.”
- Also stated in: § Data · plain
“With comprehensive experiments on 6 financial datasets, including stocks and Crypto, FinAgent significantly outperforms 12 stateof-the-art baselines in terms of 6 financial metrics with over 36% average improvement on profit.”
Quote in the paperSubstantial claim- Also stated in: § Data · plain
Probably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataThe dataset includes 398 days of asset price data with open, high, low, close, and adjusted close.
“Asset Price | | 398 × (open, high, low, close, adj_close)”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataThe dataset includes 398 days of visual data with Kline Chart and Trading Chart.
“Visual Data | | 398 × (Kline Chart, Trading Chart)”
Quote in the paperNot a substantial claimMay combine two claims.
FindingreasoningBy taking a short position, FinAgent hedged against potential trading losses and generated high returns.
“By taking a short position, it can effectively hedge against potential trading losses and generate high returns.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataLLM extracted embeddings show a clear distinction between different retrieval types, proving the effectiveness of diversified retrieval.
“We perform t-SNE visualization of its LLM extracted embedding, and we can find that the LLM extracted embedding has a clear distinction between different retrieval types, which proves the effectiveness of our method.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataAdding low-level reflection to M+ML increases ARR% by 45% to 101% for TSLA and ETHUSD, and cuts risk by 14% to 44%.
“When compared to using solely M and ML, the integration of the low-level reflection module leads to an impressive increase in ARR% by 45% to 101% for TSLA, and ETHUSD, and cutting risk by 14% to 44%.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataThe solely T method has 16% ARR% versus B&H's 29% ARR% in ETHUSD.
“As shown in Table 4 and Table 5, the 16% ARR% for solely T method starkly contrasts with the 29% ARR% of B&H in ETHUSD, highlighting the inefficacy of the stock-specific rule-based methods for cryptocurrencies and demonstrating that introducing to FinAgent significantly affects performance.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataEmploying a generalized auxiliary agent for cryptocurrency could potentially increase returns to 54%, compared to the current 44%.
“Further insights from the ablation study section for FinAgent reveal that employing a generalized auxiliary agent for cryptocurrency could potentially increase returns to 54%, compared to the current 44%.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataFinAgent significantly outperforms existing baselines in profitability, setting a new benchmark in the field.
“Table 4 and Figure 4 demonstrate our method significantly outperforms existing baselines, especially remarkable improvements in profitability, and setting a new benchmark in the field.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Findingown dataThe dataset covers 398 trading days from 2022-06-01 to 2024-01-01.
“Trading Date | From 2022-06-01 to 2024-01-01 (398 trading days)”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown dataOn the TSLA dataset, FinAgent achieves 84% and 118% improvement in ARR% and SR, respectively, over baselines.
“Notably, its performance on the TSLA dataset stands out even more, achieving 84% and 118% improvement, significantly outperforming all other baselines.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataAdding high-level reflection to ML significantly enhances ARR% and SR and reduces risk.
“When comparing the ML and MLH, the addition of the high-level reflection module significantly enhances the ARR% and SR, while notably reducing risk.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataOn the five stocks, FinAgent's ARR% and SR improvements are at least 10% and 19% over the best-performing baseline, respectively.
“FinAgent's performance on the five stocks, as measured by ARR% and SR, with enhancements of at least 10% and 19%, compared to the best-performing baseline, respectively.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataFinMem's ARR% on the AMZN dataset is 40%, underperforming the market's Buy & Hold (B&H) strategy at 42%.
“In contrast, FinMem falls short on the AMZN dataset, where its ARR% is 40%, underperforming the market's Buy & Hold (B&H) strategy at 42%.”
Quote in the paperSubstantial claimMay combine two claims.
FindingreasoningRule-based methods are optimal in controlling risk but not outstanding in capturing returns.
“We can also observe that rule-based methods are optimal in controlling risk, but not outstanding in capturing returns.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Interpretationnone givenFinAgent is the first advanced multimodal foundation agent designed for financial trading tasks.
“Notably, FinAgent is the first advanced multimodal foundation agent designed for financial trading tasks.”
- Also stated in: § Data · plain
“Notably, FinAgent is the first advanced multimodal foundation agent designed for financial trading tasks.”
Quote in the paperNot a substantial claim- Also stated in: § Data · plain
May combine two claims.
InterpretationreasoningFinAgent effectively identifies correlations between current market conditions and past market patterns and trends and integrates market information to make final and effective decisions.
“With its multi-perspective and diverse retrieval approach, FinAgent effectively identifies correlations between current market conditions and past market patterns and trends and integrates market information to make final and effective decisions.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim. May combine two claims.
InterpretationreasoningFinAgent's high-level reflection module assesses each trading decision's correctness and recommends improvements or corrective actions.
“The initial phase assesses each trading decision's correctness, identifying successes and mistakes.Subsequently, the module recommends improvements or corrective actions tailored to each identified mistake or success, fostering a continuous learning process.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S4 (based).
Interpretationnone givenFinAgent's dual-level reflection module incorporates a diversified memory retrieval system that enhances the agent's ability to learn from historical data and improve decision-making.
“Its unique dual-level reflection module not only enables rapid adaptation to market dynamics but also incorporates a diversified memory retrieval system, enhancing the agent's ability to learn from historical data and improve decision-making processes.”
Quote in the paperNot a substantial claim- Rests on an assumption: Imprecise information retrieval is a challenge for financial trading.
Classified as probably not a substantial claim. May combine two claims.
InterpretationreasoningRule-based methods are robust to outliers and noise in the data, reducing decision risk.
“This is because rule-based model methods are robust to outliers and noise in the data and thus can reduce decision risk.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningFinAgent's low-level reflection module generates detailed analysis for short-term, medium-term, and long-term perspectives.
“After taking in the price change data, the module generates detailed analysis for varying temporal horizons, spanning shortterm, medium-term to long-term perspectives.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based).
Interpretationnone givenFinAgent's low-level reflection module establishes the inherent correlation between market intelligence and price changes.
“The low-level reflection module ( §4.3) establishes the inherent correlation between market intelligence and price changes.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S4 (based).
Interpretationnone givenFinAgent's dual-level reflection module enables rapid adaptation to market dynamics.
“Its unique dual-level reflection module not only enables rapid adaptation to market dynamics but also incorporates a diversified memory retrieval system, enhancing the agent's ability to learn from historical data and improve decision-making processes.”
Quote in the paperNot a substantial claim- Rests on an assumption: Adaptability in rapidly evolving markets is a challenge for financial trading.
Probably not a substantial claim: may not pass S3 (load-bearing).
InterpretationreasoningFinAgent's high-level reflection module reflects on market conditions, price changes, and past trading decisions to derive insights and identify potential improvements in profitability.
“And the high-level reflection module ( §4.3) involves reflecting on market conditions, price changes, and other factors in the context of outcomes from past trading decisions, which aims to derive insights from previous experiences and identify potential improvement in profitability by assessing the efficacy of historical decisions and offering recommendations for future decision-making processes.”
Quote in the paperNot a substantial claimClassified as probably not a substantial claim.
InterpretationreasoningFinAgent's diversified memory retrieval system separates trading and retrieval tasks to enhance focus on their specific functions and minimize noise in results.
“FinAgent introduces a diversified memory retrieval system for the market intelligence and reflection modules, separating trading and retrieval tasks to enhance focus on their specific functions and minimize noise in the results (Ch3).”
Quote in the paperSubstantial claim- Rests on an assumption: Imprecise information retrieval is a challenge for financial trading.
May combine two claims.
PremisecitationReinforcement learning-based systems demonstrate enhanced adaptability but encounter substantial obstacles, such as the need for extensive training data and the inexplainability of decision-making processes.
“Reinforcement learningbased systems [2] demonstrate enhanced adaptability but encounter substantial obstacles, such as the need for extensive training data and the inexplainability of decision-making processes.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [2] 2. Bo An, Shuo Sun, and Rundong Wang. 2022. Deep reinforcement learning for quantitative trading: Challenges and opportunities. IEEE Intelligent Systems 37, 2 (2022), 23-26. source
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Interpretationnone givenFinAgent generates summaries for trading tasks and provides query fields for retrieval tasks, with query texts including different retrieval types tailored for focused retrieval.
“We not only generate summaries for trading tasks but also provide query fields for retrieval tasks.These query texts include different retrieval types, tailored to enable focused retrieval of specific types of information.”
Quote in the paperNot a substantial claim- Rests on an assumption: Imprecise information retrieval is a challenge for financial trading.
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
PremisecitationFMP API provides data about stock historical price and news, company financial statements, and cryptocurrencies.
“FMP API provides data about stock historical price and news, company financial statements, and cryptocurrencies.Entry is https://site.financialmodelingprep.com.”
Quote in the paperNot a substantial claim0 of 1 sources checked- cites https://site.financialmodelingprep.com
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
May combine two claims.
InterpretationreasoningFinAgent has superior stability and robustness compared to other baselines.
“This underscores the superior stability and robustness of FinAgent compared to other baselines.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
InterpretationreasoningFinAgent's tool-augmented decision-making module integrates information, augmented tools, and trader preferences to make final investment decisions with comprehensive analysis.
“The tool-augmented decision-making module ( §4.4) integrates the aforementioned information, along with augmented tools and trader preferences, to make final investment decisions with a comprehensive analysis.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.
Interpretationnone givenFinAgent integrates established trading strategies and expert insights, ensuring its trading approaches are data-driven and rooted in sound financial principles.
“Moreover, FinAgent integrates established trading strategies and expert insights, ensuring that its trading approaches are both data-driven and rooted in sound financial principles.”
Quote in the paperNot a substantial claim- Rests on an assumption: Integration of domain knowledge is a challenge for financial trading.
May combine two claims.
Interpretationnone givenFinAgent exhibits high reasoning ability and generalizability.
“This paper introduces FinAgent, a financial trading agent powered by LLM that exhibits high reasoning ability and generalizability.”
Quote in the paperNot a substantial claim- Rests on an assumption: The selected five stocks and Ethereum are representative of diverse financial environments and provide a robust basis for assessing FinAgent's generalization capabilities.
Classified as probably not a substantial claim.
Study design and conductnone givenAll benchmarks are conducted within the same RL environment for both training and evaluation to ensure equitable comparison.
“To ensure equitable comparison, all benchmarks are conducted within the same RL environment for both training and evaluation.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Interpretationnone givenFinAgent's market intelligence module processes multimodal data to provide precise analysis of financial market trends, offering insights for future trading tasks.
“Specifically, FinAgent's market intelligence module processes multimodal data, such as numerical, textual, and visual, to provide precise analysis of financial market trends, offering insights for future trading tasks (Ch1).”
Quote in the paperSubstantial claim- Rests on an assumption: Insufficient multimodal data processing ability is a challenge for financial trading.
Classified as probably not a substantial claim.
InterpretationreasoningThe risk compromise relates to the chosen investor preference of an aggressive trader.
“This result relates to our chosen investor preference of an aggressive trader.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing).
Interpretationnone givenFinAgent's memory module supports the market intelligence, low-level reflection, and high-level reflection modules by offering storage capabilities and vector retrieve functions.
“The primary role of the memory module ( §4.2) is to support the aforementioned three modules by offering storage capabilities and vector retrieve functions.”
Quote in the paperNot a substantial claimProbably not a substantial claim: may not pass S4 (based). May combine two claims.
Interpretationnone givenFinAgent's dual-level reflection module is capable of rapidly adapting to market dynamics and enhancing the agent's ability to learn from historical data and improve its decision-making process.
“A uniquely designed dual-level reflection module is developed, capable of not only rapidly adapting to market dynamics but also enhancing the agent's ability to learn from historical data and improve its decision-making process (Ch2).”
Quote in the paperNot a substantial claim- Rests on an assumption: Adaptability in rapidly evolving markets is a challenge for financial trading.
Probably not a substantial claim: may not pass S4 (based).
Interpretationnone givenFinAgent's market intelligence module processes numerical, textual, and visual data to accurately analyze the financial market.
“FinAgent's market intelligence module processes a diverse range of data-numerical, textual, and visual-to accurately analyze the financial market.”
Quote in the paperNot a substantial claim- Rests on an assumption: Insufficient multimodal data processing ability is a challenge for financial trading.
Classified as probably not a substantial claim. May combine two claims.
Study design and conductnone givenPast and future asset price increases and decreases are visible during training, but only past trends are visible during testing to prevent data leakage.
“It is important to note that past and future asset price increases and decreases are visible during the training phase, but only past trends are visible during the testing phase to prevent data leakage.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Interpretationnone givenFinAgent's dual-level reflection module combines low-level reflection that analyzes market price movement for insights and high-level reflection that assesses past trading decisions for improvement.
“Our duel-level reflection module combines a low-level reflection that analyzes market price movement for insights, while the highlevel reflection assesses past trading decisions for improvement, emulating the learning process in decision-making.”
Quote in the paperNot a substantial claim- Rests on an assumption: Adaptability in rapidly evolving markets is a challenge for financial trading.
The paper relies on this without support.
Advanced AI techniques in financial trading often face challenges due to inadequate handling of multimodal data and limited generalizability across various tasks.
“While advanced AI techniques like deep learning and reinforcement learning are extensively utilized in finance, their application in financial trading tasks often faces challenges due to inadequate handling of multimodal data and limited generalizability across various tasks.”
The paper relies on this without support.
Rule-based trading systems are rigid and struggle to adapt to market volatility, often resulting in underperformance in evolving markets.
“Rule-based trading systems are rigid and struggle to adapt to market volatility, often resulting in underperformance in evolving markets.”
The paper relies on this without support.
Reinforcement learning-based systems struggle with generalizing across diverse market conditions, are sensitive to market noise, and often fail to integrate multimodal market intelligence like news and reports into their analysis.
“Additionally, they struggle with generalizing across diverse market conditions, are sensitive to market noise, and often fail to integrate multimodal market intelligence like news and reports into their analysis.”
The paper relies on this without support.
There is a natural gap between QA tasks and sequential decision-making in trading.
“However, there is a natural gap between QA tasks and sequential decision-making in trading.”
The paper relies on this without support.
Insufficient multimodal data processing ability is a challenge for financial trading.
“• Ch1: Insufficient Multimodal Data Processing Ability.Processing numerical, textual, and visual market intelligence data significantly requires advanced analytical methods to extract key insights and predict market trends.”
The paper relies on this without support.
Imprecise information retrieval is a challenge for financial trading.
“• Ch2: Imprecise information retrieval.Mixing retrieval with main tasks and relying on brief summaries causes imprecise searches, introducing irrelevant data and reducing performance.”
The paper relies on this without support.
Adaptability in rapidly evolving markets is a challenge for financial trading.
“• Ch3: Adaptability in Rapidly Evolving Markets.Financial trading requires the ability to quickly adapt to fluctuating market conditions.Traditional methods often fall short, highlighting the necessity for models capable of responding to real-time data and adjusting strategies according to historical market trends.”
The paper relies on this without support.
Integration of domain knowledge is a challenge for financial trading.
“• Ch4: Integration of Domain Knowledge.Current models often struggle to integrate established methods such as expert guidance and advanced trading tools effectively, leading to a decline in both the effectiveness and depth of market analysis.”
The paper relies on this without support.
Reasoning for actions is a challenge for financial trading.
“• Ch5.Reasoning for Actions.The black-box nature of many sophisticated AI models, directly giving results of decisions without providing the reasoning process.”
The paper relies on this without support.
High returns often come with high risks.
“It is worth noting that high returns often come with high risks.”
The paper relies on this without support.
Cryptocurrencies have higher trading frequency than stocks.
“not for cryptocurrencies with higher trading frequency.”
The paper relies on this without support.
The selected five stocks and Ethereum are representative of diverse financial environments and provide a robust basis for assessing FinAgent's generalization capabilities.
“Chosen for their extensive news coverage and representation of different market sectors, these data provide a robust basis for assessing FinAgent's generalization capabilities across diverse financial environments.”
The paper relies on this without support.
Market intelligence can be categorized as positive, negative, or neutral based on its impact on market perceptions and potential outcomes.
“Market intelligence can be categorized as positive, negative, or neutral based on its impact on market perceptions and potential outcomes.”
The paper relies on this without support.
Some news articles report the percentage increase or decrease in a company's stock price after an event occurs.
“Some news will detail the percentage increase or decrease in a company's stock price after some event occurs.”
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?
In [31, 53, 57], rule based strategies such as 'Buy and Hold', 'Mean Reversion' and 'ShortTerm Reversal' are used as baseline.
“In [31, 53, 57], rule based strategy such as”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
Section 5.3 (Baselines; page unavailable) says the paper uses rule-based trading strategies as baselines and explicitly names Buy-and-Hold. It lists MACD, KDJ&RSI, and ZMR as the other strategies; our copy does not identify Mean Reversion or ShortTerm Reversal as baselines. (LLM, conf=0.85)
Experimental — accuracy not yet measuredFinAgent is the first multimodal agent with layered memory and layered reflection design, with an additional multimodal module that takes in numeric, text and image data.
“FinAgent[57] proposed the first multimodal agent with similar layered memory and layered reflection design, with an additional multimodal module that takes in numeric, text and image data.”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
The Data section describes FinAgent’s numerical, textual, and visual inputs and its dual-level reflection; §4.2 describes three memory components. The paper’s first-of-its-kind claim is narrower: it calls FinAgent the first advanced multimodal foundation agent designed for financial trading, not the first multimodal agent generally. It describes distinct memory components, but does not clearly characterize their design as layered. (LLM, conf=0.87)
Experimental — accuracy not yet measuredFinAgent demonstrated significant trading performance improvements over FinMem, which uses a similar architecture but without visual input.
“It demonstrated significant trading performance improvements over FinMem which uses a similar architecture but without visual input.”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
The Experimental Results section reports that FinAgent significantly outperformed the baselines, including FinMem, on the tested stocks. It also says FinAgent had slightly lower returns than FinMem on ETH. The Related Work and Baselines sections describe FinMem as an LLM trading agent, but do not establish that it has a similar architecture or lacks visual input. (LLM, conf=0.85)
Experimental — accuracy not yet measuredFinAgent demonstrated superior performance in backtesting compared to other agents including FinMem.
“This framework has demonstrated superior performance in backtesting compared to other agents including FinMem.”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
Robust memory is essential for practical applications such as personalized chatbots and financial analysis, and as a key enabler toward AGI.
“Robust memory is essential not only for practical applications such as personalized chatbots (Li et al., 2025a) and financial analysis (Zhang et al., 2024), but also as a key enabler toward Artificial General Intelligence (AGI) (Fang et al., 2025a).”
From RealMem: Benchmarking LLMs in Real-World Memory-Driven Interaction (2026) · cites this paper as Zhang et al., 2024
Financial reports are extensively utilized by financial trading agents like FinMem, TradingGPT, and FinAgent to enrich the agents' memory and make informed trading decisions.
“They are extensively utilized by financial trading agents like FinMem [53], TradingGPT [27], and FinAgent [57]. These works incorporate financial reports to enrich the agents' memory and make informed trading decisions.”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
News data from reputable sources is extensively used in various studies to stay up-to-date with the real-world financial market.
“This type of data is extensively used in various studies [20, 27, 51, 53, 57] to stay up-to-date with the real-world financial market.”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
FinAgent incorporates expert guidance from SeekingAlpha as a crucial input to its decision-making module.
“For example, FinAgent [57] incorporates expert guidance from SeekingAlpha 1 as a crucial input to its decision-making module.”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
Only FinAgent [57] extends its backtesting on cryptocurrency market, specifically trading ETH [4].
“Only FinAgent [57] extends its backtesting on the cryptocurrency market, specifically trading ETH [4].”
From Large Language Model Agent in Financial Trading: A Survey (2026) · cites this paper as [57]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Baseline count changes from twelve to nine
MinorThe methods section specifies four rule-based strategies and eight advanced algorithms, totaling twelve baselines, but the results section says it compares against nine.
“We compared FinAgent with 9 baseline methods in terms of 6 financial metrics.”
“We compare and evaluate the trading performance of FinAgent with four widely accepted conventional rule-based trading strategies (B&H, MACD, KDJ&RSI and ZMR) and eight advanced algorithms.”
Appendix adds an unlisted seventh asset
MinorThe main dataset description identifies six assets-five stocks and Ethereum-but the appendix also says the portfolio contains a foreign exchange pair.
“We selected a varied portfolio comprising five stocks Apple Inc. (AAPL), Amazon.comInc. (AMZN), Alphabet Inc. (GOOGL), Microsoft Corporation (MSFT), and Tesla Inc. (TSLA), a foreign exchange pair, and a prominent cryptocurrency, Ethereum (ETH).”
“Our diversified portfolio includes five major stocks: Apple Inc. (AAPL), Amazon.comInc. (AMZN), Alphabet Inc. (GOOGL), Microsoft Corporation (MSFT), and Tesla Inc. (TSLA) and a prominent cryptocurrency named Ethereum (ETHUSD).”
Cites
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 · see citing works
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 · see citing works
Larian Studios Dev Team (2024) · doi:10.1145/3292500.3330701 · Cited by 8,618 · see citing works
LLaMA: Open and Efficient Foundation Language Models
Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozière, B., Goyal, N., Hambro, E., Azhar, F., et al. (2023) · arXiv · doi:10.48550/arxiv.2302.13971 · Cited by 4,602 · see citing works
Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question AnsweringPreprint
Lu, P., Mishra, S., Xia, T., et al. (2022) · Neural Information Processing Systems · doi:10.48550/arxiv.2209.09513 · Cited by 2,754 · see citing works
PaLM: Scaling Language Modeling with Pathways
Chowdhery, A., Narang, S., Devlin, J., Bosma, M., Mishra, G., Roberts, A., Barham, P., Chung, H. W., Sutton, C., Gehrmann, S., et al. (2022) · arXiv · doi:10.48550/arxiv.2204.02311 · Cited by 2,590 · cited 1× · see citing works
“Financial trading systems, developed from technical analysis strategies [9], enhance these markets by enabling efficient trading.”
Self-citationGenerative 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 · see citing works
Soft Actor-Critic Algorithms and Applications
Haarnoja, T., Zhou, A., Hartikainen, K., Tucker, G., Ha, S., Tan, J., Kumar, V., Zhu, H., Gupta, A., Abbeel, P., et al. (2018) · arXiv · doi:10.48550/arxiv.1812.05905 · Cited by 1,994 · see citing works
Toolformer: Language Models Can Teach Themselves to Use Tools
Schick, T., Dwivedi-Yu, J., Dessì, R., Raileanu, R., Lomeli, M., Zettlemoyer, L., Cancedda, N., & Scialom, T. (2023) · arXiv · doi:10.48550/arxiv.2302.04761 · Cited by 1,150 · see citing works
Stock market's price movement prediction with LSTM neural networks
Nelson, D. M. Q., Pereira, A. C. M., & de Oliveira, R. A. (2017) · 2017 International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn.2017.7966019 · Cited by 951 · cited 1× · see citing works
“The field of artificial intelligence and natural language processing has reached a significant milestone with the emergence of LLMs 25].”
Deep Direct Reinforcement Learning for Financial Signal Representation and Trading
Deng, Y., Bao, F., Kong, Y., Ren, Z., & Dai, Q. (2017) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2016.2522401 · Cited by 822 · cited 1× · see citing works
“Later on, we provide the formal formulation of FinAgent , which integrates LLMs into the RL pipeline to enable flexible reasoning and decision-making in financial trading.”
LaMDA: Language Models for Dialog Applications
Thoppilan, R., De Freitas, D., Hall, J., Shazeer, N., Kulshreshtha, A., Cheng, H.-T., Jin, A., Bos, T., Baker, L., Du, Y., et al. (2022) · arXiv · doi:10.48550/arxiv.2201.08239 · Cited by 778 · see citing works
MM-REACT: Prompting ChatGPT for Multimodal Reasoning and ActionPreprint
Yang, Z.-Y., Li, L., Wang, J., Lin, K., Azarnasab, E., Ahmed, F., Liu, Z.-C., Liu, C., Zeng, M., & Wang, L.-J. (2023) · arXiv.org · doi:10.48550/arxiv.2303.11381 · Cited by 660 · see citing works
Temporal Relational Ranking for Stock Prediction
Feng, F., He, X., Wang, X., Luo, C., Liu, Y., & Chua, T.-S. (2019) · ACM Transactions on Information Systems · doi:10.1145/3309547 · Cited by 553 · cited 1× · see citing works
“For instance, Feng et al. [10] enhance graph convolutional networks (GCNs) with temporal convolutions for mining inter-stock relations.”
Self-citationChameleon: Plug-and-Play Compositional Reasoning with Large Language ModelsPreprint
Lu, P., Peng, B., Cheng, H., Galley, M., Chang, K.-W., Wu, Y., Zhu, S.-C., & Gao, J.-F. (2023) · Neural Information Processing Systems · doi:10.48550/arxiv.2304.09842 · Cited by 534 · see citing works
Stock Movement Prediction from Tweets and Historical Prices
Xu, Y., & Cohen, S. B. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1183 · Cited by 434 · cited 1× · see citing works
“Initiatives like AutoGPT [51] and MetaGPT [12], Voyager [43], and AI agents [28, 35] expand LLMs’ capabilities to complex tasks involving reasoning and collaboration, significantly advancing technology and impacting daily life.”
Learn to Explain: Multimodal Reasoning Via Thought Chains for Science Question Answering
Lu, P., Mishra, S. R., Xia, T., Qiu, L., Chang, K.-W., Zhu, S. C., Tafjord, O., Clark, P., & Kalyan, A. (2022) · neural information processing systems · doi:10.52202/068431-0182 · Cited by 420 · see citing works
BloombergGPT: A Large Language Model for Finance
Wu, S., Irsoy, O., Lu, S., Dabravolski, V., Dredze, M., Gehrmann, S., Kambadur, P., Rosenberg, D., & Mann, G. (2023) · arXiv · Cited by 357 · see citing works
FinGPT: Open-Source Financial Large Language Models
Yang, H., Liu, X.-Y., & Wang, C. D. (2023) · arXiv · doi:10.48550/arxiv.2306.06031 · Cited by 326 · see citing works
SciBench: Evaluating College-Level Scientific Problem-Solving Abilities of Large Language ModelsPreprint
Wang, X.-X., Hu, Z.-N., Lu, P., Zhu, Y., Zhang, J., Subramaniam, S., Loomba, A. R., Zhang, S., Sun, Y., & Wang, W. (2023) · International Conference on Machine Learning · doi:10.48550/arxiv.2307.10635 · Cited by 300 · see citing works
Self-citationHu, Z., Liu, W., Bian, J., Liu, X., & Liu, T. (2018) · ACM International Conference on Web Search and Data Mining (WSDM) · doi:10.1145/3159652.3159690 · Cited by 281 · cited 1× · see citing works
“FinMEM [56] presents an LLM agent with a human-aligned memory mechanism and character design for automated trading.”
TheoremQA: A Theorem-driven Question Answering datasetPreprint
Chen, W., Yin, M., Ku, M. W. F., Wan, Y., Ma, X., Xu, J.-Y., Xia, T., Wang, X., & Lu, P. (2023) · Conference on Empirical Methods in Natural Language Processing · doi:10.48550/arxiv.2305.12524 · Cited by 278 · see citing works
Self-citationWebGPT: Browser-assisted question-answering with human feedback
Nakano, R., Hilton, J., Balaji, S., Wu, J., Ouyang, L., Kim, C., Hesse, C., Jain, S., Kosaraju, V., Saunders, W., et al. (2021) · arXiv · doi:10.48550/arxiv.2112.09332 · Cited by 272 · see citing works
Technical Analysis of Stock Trends
Edwards, R. D., Magee, J. C., & Bassetti, W. H. C. (2018) · CRC Press eBooks · doi:10.4324/9781315115719 · Cited by 267 · see citing works
Voyager: An Open-Ended Embodied Agent with Large Language Models
Wang, G., Xie, Y., Jiang, Y., Mandlekar, A., Xiao, C., Zhu, Y., Fan, L., & Anandkumar, A. (2023) · arXiv · doi:10.48550/arxiv.2305.16291 · Cited by 222 · see citing works
MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework
Hong, S., Zhuge, M., Chen, J., Zheng, X., Cheng, Y., Zhang, C., Wang, J., Wang, Z., Yau, S. K. S., Lin, Z., et al. (2023) · arXiv · doi:10.48550/arxiv.2308.00352 · Cited by 214 · see citing works
SciEval: A Multi-Level Large Language Model Evaluation Benchmark for Scientific ResearchPreprint
Sun, L., Han, Y., Zhao, Z., Ma, D., Shen, Z.-W., Chen, B.-C., Chen, L., & Yu, K. (2023) · AAAI Conference on Artificial Intelligence · doi:10.48550/arxiv.2308.13149 · Cited by 187 · see citing works
Reinforcement-Learning Based Portfolio Management with Augmented Asset Movement Prediction States
Ye, Y., Pei, H., Wang, B., Chen, P., Zhu, Y., Xiao, J., & Li, B. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i01.5462 · Cited by 180 · cited 1× · see citing works
“Later on, we provide the formal formulation of FinAgent , which integrates LLMs into the RL pipeline to enable flexible reasoning and decision-making in financial trading.”
OpenAI, :, Hurst, A., Lerer, A., Goucher, A. P., Perelman, A., Ramesh, A., Clark, A., Ostrow, A., Welihinda, A., et al. (2024) · arXiv · Cited by 180 · see citing works
Sawhney, R., Agarwal, S., Wadhwa, A., & Shah, R. R. (2020) · Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.676 · Cited by 146 · cited 1× · see citing works
“Sawhney et al. [30] focus on stock industry data and links between company CEOs.”
Wang, Z., Huang, B., Tu, S., Zhang, K., & Xu, L. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i1.16144 · Cited by 142 · see citing works
Adaptive Quantitative Trading: An Imitative Deep Reinforcement Learning Approach
Liu, Y., Liu, Q., Zhao, H., Pan, Z., & Liu, C. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i02.5587 · Cited by 141 · cited 2× · see citing works
“The reward function leverages the change of market capitals (earned/lost money) [16] with consideration of commission fee [38, 45].”
Generate rather than Retrieve: Large Language Models are Strong Context Generators
Yu, W., Iter, D., Wang, S., Xu, Y., Ju, M., Sanyal, S., Zhu, C., Zeng, M., & Jiang, M. (2022) · arXiv · doi:10.48550/arxiv.2209.10063 · Cited by 95 · see citing works
FinRL: A Deep Reinforcement Learning Library for Automated Stock Trading in Quantitative Finance
Liu, X.-Y. (2020) · SSRN Electronic Journal · doi:10.2139/ssrn.3737257 · Cited by 90 · see citing works
Liu, X.-Y., Yang, H., Gao, J., & Wang, C. D. (2021) · Proceedings of the Second ACM International Conference on AI in Finance · doi:10.1145/3490354.3494366 · Cited by 87 · see citing works
Chen, C., Zhao, L., Bian, J., Xing, C., & Liu, T.-Y. (2019) · Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining · doi:10.1145/3292500.3330663 · Cited by 73 · cited 2× · see citing works
“For instance, if a past product launch significantly boosted a company’s stock, a recent launch might have a similar effect 3 .”
Self-citationWang, J., Sun, T., Liu, B., Cao, Y., & Zhu, H. (2019) · Proceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2019/514 · Cited by 73 · cited 1× · see citing works
“The field of artificial intelligence and natural language processing has reached a significant milestone with the emergence of LLMs 25].”
Cognitive Architectures for Language Agents
Sumers, T. R., Yao, S., Narasimhan, K., & Griffiths, T. L. (2023) · arXiv · doi:10.48550/arxiv.2309.02427 · Cited by 72 · see citing works
Technical Analysis of Stock Trends
Hooper, C. M., Thompson, C., Laver-Bradbury, C., & Gale, C. K. (2012) · CRC Press eBooks · doi:10.1201/b14301 · Cited by 61 · see citing works
Commission Fee is not Enough: A Hierarchical Reinforced Framework for Portfolio Management
Wang, R., Wei, H., An, B., Feng, Z., & Yao, J. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i1.16142 · Cited by 45 · cited 1× · see citing works
“Later on, we provide the formal formulation of FinAgent , which integrates LLMs into the RL pipeline to enable flexible reasoning and decision-making in financial trading.”
Self-citationDynamic Prompt Learning via Policy Gradient for Semi-structured Mathematical Reasoning
Lu, P., Qiu, L., Chang, K.-W., Wu, Y. N., Zhu, S.-C., Rajpurohit, T., Clark, P., & Kalyan, A. (2022) · arXiv · doi:10.48550/arxiv.2209.14610 · Cited by 45 · see citing works
Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions
Yang, H., Yue, S., & He, Y. (2023) · arXiv · doi:10.48550/arxiv.2306.02224 · Cited by 44 · see citing works
Sawhney, R., Khanna, P., Aggarwal, A., Jain, T., Mathur, P., & Shah, R. R. (2020) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2020.emnlp-main.643 · Cited by 37 · cited 1× · see citing works
“Other data sources such as economics news [13] and earning calls [31] are also used to improve the prediction performance.”
TALM: Tool Augmented Language Models
Parisi, A., Zhao, Y., & Fiedel, N. (2022) · arXiv · doi:10.48550/arxiv.2205.12255 · Cited by 35 · see citing works
Self-citationQlib: An AI-oriented Quantitative Investment PlatformPreprint
Yang, X., Liu, W., Zhou, D., et al. (2020) · RePEc: Research Papers in Economics · doi:10.48550/arxiv.2009.11189 · Cited by 29 · see citing works
EarnHFT: Efficient Hierarchical Reinforcement Learning for High Frequency Trading
Qin, M., Sun, S., Zhang, W., Xia, H., Wang, X., & An, B. (2023) · arXiv · doi:10.48550/arxiv.2309.12891 · Cited by 29 · see citing works
Self-citationDing, Y. C., Liu, W., Bian, J., Zhang, D., & Liu, T. (2018) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3219819.3220113 · Cited by 23 · see citing works
Quantitative Day Trading from Natural Language using Reinforcement Learning
Sawhney, R., Wadhwa, A., Agarwal, S., & Shah, R. R. (2021) · Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT) · doi:10.18653/v1/2021.naacl-main.316 · Cited by 23 · cited 1× · see citing works
“A state represents RL agents’ perception on the financial market based on price information, limited order book [29], technical indicators, trend prediction [54], financial news [32], experts’ investment behaviors [8] and overall market status [47].”
Deep Reinforcement Learning for Quantitative Trading: Challenges and Opportunities
An, B., Sun, S., & Wang, R. (2022) · IEEE Intelligent Systems · doi:10.1109/mis.2022.3165994 · Cited by 22 · see citing works
Self-citationMastering Stock Markets with Efficient Mixture of Diversified Trading Experts
Sun, S., Wang, X., Xue, W., Lou, X., & An, B. (2023) · Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining · doi:10.1145/3580305.3599424 · Cited by 16 · cited 1× · see citing works
“Recently, there has been growing interest in enhancing LLM agents with external tools and modular methods as AI agents.”
Self-citationMathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts
Lu, P., Bansal, H., Xia, T., Liu, J., Li, C., Hajishirzi, H., Cheng, H., Chang, K.-W., Galley, M., & Gao, J. (2023) · arXiv · doi:10.48550/arxiv.2310.02255 · Cited by 12 · see citing works
AppAgent: Multimodal Agents as Smartphone Users
Zhang, C., Yang, Z., Liu, J., Li, Y., Han, Y., Chen, X., Huang, Z., Fu, B., & Yu, G. (2023) · arXiv · doi:10.48550/arxiv.2312.13771 · Cited by 11 · cited 3× · see citing works
“Notably, FinAgent is the first advanced multimodal foundation agent designed for financial trading tasks.”
What Your Next Check-in Might Look Like: Next Check-in Behavior Prediction
Sun, H., Cao, C., Chu, X., Hu, T., Lu, J., He, L., Wang, Z., He, H., & Xiong, H. (2023) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3625234 · Cited by 10 · see citing works
FinMem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design
Yu, Y., Li, H., Chen, Z., Jiang, Y., Li, Y., Zhang, D., Liu, R., Suchow, J. W., & Khashanah, K. (2023) · arXiv · doi:10.48550/arxiv.2311.13743 · Cited by 6 · see citing works
TradeMaster: A Holistic Quantitative Trading Platform Empowered by Reinforcement Learning
Sun, S., Qin, M., Zhang, W., Xia, H., Zong, C., Ying, J., Xie, Y., Zhao, L., Wang, X., & An, B. (2023) · Advances in Neural Information Processing Systems 36 · doi:10.52202/075280-2576 · Cited by 3
Self-citationPangu-Agent: A Fine-Tunable Generalist Agent with Structured Reasoning
Christianos, F., Papoudakis, G., Zimmer, M., Coste, T., Wu, Z., Chen, J., Khandelwal, K., Doran, J., Feng, X., Liu, J., et al. (2023) · arXiv · doi:10.48550/arxiv.2312.14878 · Cited by 2 · see citing works
OpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., Aleman, F. L., Almeida, D., Altenschmidt, J., Altman, S., et al. (2023) · arXiv · doi:10.48550/arxiv.2303.08774
Flagged first, then by citations
Research works printed but not linked: 2
- 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. Advances in neural information processing systems, Vol. 30 (2017).
- Shuo Sun, Molei Qin, Xinrun Wang, and Bo An. 2023. PRUDEX-Compass: Towards systematic evaluation of reinforcement learning in financial markets. Transactions on Machine Learning Research (2023).
Cited by
Sorted by citations
Tool learning with large language models: a survey
Qu, C., Dai, S., Wei, X., Cai, H., Wang, S., Yin, D., Xu, J., & Wen, J.-R. (2025) · Frontiers of Computer Science · doi:10.1007/s11704-024-40678-2 · Cited by 148 · see citing works
A comprehensive survey on integrating large language models with knowledge-based methods
Yang, W., Some, L., Bain, M., & Kang, B. H. (2025) · Knowledge-Based Systems · doi:10.1016/j.knosys.2025.113503 · Cited by 88 · see citing works
LLM-Based Agents for Tool Learning: A Survey
Xu, W., Huang, C., Gao, S., & Shang, S. (2025) · Data Science and Engineering · doi:10.1007/s41019-025-00296-9 · Cited by 69 · see citing works
FinMem: A Performance-Enhanced LLM Trading Agent With Layered Memory and Character Design
Yu, Y., Li, H., Chen, Z., Jiang, Y., Li, Y., Suchow, J. W., Zhang, D., & Khashanah, K. (2025) · IEEE Transactions on Big Data · doi:10.1109/tbdata.2025.3593370 · Cited by 23 · see citing works
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 32 · see citing works
“Here are some of its industry use cases: LLM-powered Chatbots and virtual assistants - such as ChatGPT [17], Claude [18], Gemini [19], DeepSeek [20] - as well as automated financial trading agents [21]–[25].”
Zhang, Z., Jiang, C., & Lu, M. (2025) · Big Data and Cognitive Computing · doi:10.3390/bdcc9060161 · Cited by 14 · see citing works
A Literature Review of Gen AI Agents in Financial Applications: Models and ImplementationsPreprint
Joshi, S. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5133985 · Cited by 13 · see citing works
HedgeAgents: A Balanced-aware Multi-agent Financial Trading System
Li, X., YaWen, Z., Xing, X., et al. (2025) · The Web Conference · doi:10.1145/3701716.3715232 · Cited by 12 · see citing works
Self-citationRTBAgent: A LLM-based Agent System for Real-Time Bidding
Cai, L., He, J., Li, Y., et al. (2025) · doi:10.1145/3701716.3715259 · Cited by 11 · see citing works
Large Language Model Agent in Financial Trading: A Survey
Ding, H., Li, Y., Wang, J., Chen, H., Guo, D., & Zhang, Y. (2026) · Proceedings of the 2026 9th International Conference on Computers in Management and Business · doi:10.1145/3802463.3802475 · Cited by 10 · see citing works
Joshi, S. (2025) · doi:10.38124/ijisrt/25may964 · Cited by 10 · see citing works
“Reference [47] proposes a multimodal foundation agent for financial trading, building upon Josih's previous research [82-96].”
Stock Market Forecasting: From Traditional Predictive Models to Large Language Models
Darwish, M., Hassanien, E. E., & Eissa, A. H. B. (2025) · Computational Economics · doi:10.1007/s10614-025-11024-w · Cited by 27 · see citing works
“Zhang et al. introduced FinAgent (Zhang et al., 2024 ), the first multimodal LLM agent for financial trading that is capable of processing different data sources includ - ing numerical, textual and visual information utilizing recent LLM capabilities of processing different data types including…”
InfluentialTradingAgents: Multi-Agents LLM Financial Trading Framework
Xiao, Y., Sun, E., Luo, D., & Wang, W. (2024) · arXiv · doi:10.48550/arxiv.2412.20138 · Cited by 6 · see citing works
“Reflection-driven agents, such as FinMem (Yu et al., 2023) and FinAgent (Zhang et al., 2024b), use layered memorization and multimodal data to summarize inputs into memories, inform decisions, and incorporate technical indicators, achieving superior backtest performance while mitigating hallucinations (Ji et al., 2023…”
Building Multi-Agent Copilot towards Autonomous Agricultural Data Management and Analysis
Pan, Y., Sun, J., Yu, H., Luck, J. D., Bai, G., Chamara, N., Ge, Y., & Awada, T. N. (2024) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata62323.2024.10826038 · Cited by 6 · see citing works
“…umbrella concept which encompasses a lot of subcategories, including augmenting LLMs with external knowledge [11]–[14], augmenting LLMs with mathematical reasoning ability [15], [16], augmenting LLMs with program executors [17], [18], and augmenting LLMs with domain specific tool sets [19], [20].”
Self-citationSaha, P., Lyu, J., Saxena, A., Zhao, T., & Mehta, D. B. (2025) · ACM International Conference on AI in Finance (ICAIF) · doi:10.1145/3768292.3770387 · Cited by 8 · see citing works
“For instance, FinAgent [53] is, to our knowledge, the first financial trading agent to combine multimodal market perception, tool-augmented reasoning, diversified memory retrieval, and a dual-level reflection mechanism.”
InfluentialChoi, J. Y., & Yoo, T. K. (2026) · Discover Health Systems · doi:10.1007/s44250-026-00343-w · Cited by 10 · see citing works
“Beyond medicine, related agentic and integration approaches are emerging in finance, urban systems, and logistics, suggesting a broader shift from passive prediction to context-aware, goal-directed agents [12–14].”
Wang, Z., Moriyama, S., Wang, W.-Y., Gangopadhyay, B., & Takamatsu, S. (2026) · IEEE Transactions on Artificial Intelligence · doi:10.1109/tai.2026.3676025 · Cited by 4 · see citing works
Your AI, Not Your View: The Bias of LLMs in Investment Analysis
Lee, H., Seo, J., Park, S., Lee, J., Ahn, W., Choi, C., Lopez-Lira, A., & Lee, Y. (2025) · ACM International Conference on AI in Finance (ICAIF) · doi:10.1145/3768292.3770375 · Cited by 3 · see citing works
When DeepSeek-R1 meets financial applications: benchmarking, opportunities, and limitations
Liu, S., Chen, L., Yan, J., Jiang, Y., Wang, X., Li, X., & Yang, Q. (2025) · Frontiers of Information Technology & Electronic Engineering · doi:10.1631/fitee.2500227 · Cited by 2 · see citing works
Self-citationSun, Y., Wu, H., Zhu, C., Si, Y., Chen, Q., Zhang, Y., Zhang, K., Li, J., Cai, J., Wang, Y., et al. (2025) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2025.3584857 · Cited by 2 · see citing works
“These models demonstrate broad capabilities and serve as foundational tools, offering novel solutions to complex problems across diverse sectors such as finance [23], [24], and medicine [25], [26].”
Rezaei, H., Rabhi, F., & Beheshti, A. (2025) · IEEE International Conference on Web Services (ICWS) · doi:10.1109/icws67624.2025.00136 · Cited by 2 · see citing works
Han, S., Zhang, J., Shen, Y., Yan, K., & Li, H. (2025) · Frontiers of Information Technology & Electronic Engineering · doi:10.1631/fitee.2500414 · Cited by 6 · see citing works
Language Model Guided Reinforcement Learning in Quantitative Trading
Darmanin, A., & Vella, V. (2025) · International Conference on Foundation and Large Language Models (FLLM) · doi:10.1109/fllm67465.2025.11390898 · Cited by 3 · see citing works
“The final augmented baseline prompt, generalized across equities and regimes through this process, was then subjected to three additional refinements.”
Towards Co-designing a Human-Agent Interface: A Case Study in Online Grooming Detection.
Daish, P., Micallef, N., Lorenzo-Dus, N., Paiement, A., & Sahoo, D. R. (2024) · International Conference on Human-Agent Interaction (HAI) · doi:10.1145/3687272.3690893 · Cited by 2 · see citing works
“The integration of artificial intelligence (AI) agents into professional workflows is growing [6, 10, 22, 23].”
A Systematic Analysis Approach of AI's Role in Enhancing Financial Decision-Making
Maaqili, Y. E., & Rhouiri, M. (2025) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3693-9684-1.ch013 · Cited by 2 · see citing works
Luo, C. S., He, H. (., Cucuringu, M., & Ma, T. (2025) · Quantitative Finance · doi:10.1080/14697688.2025.2548897 · Cited by 2 · see citing works
“All data is sourced from Financial Modeling Prep (FMP),¶ an online financial data service which has been utilized in previous studies (Liu et al. 2023, Zhang et al. 2024).”
Automated Trading Framework Using LLM-Driven Features and Deep Reinforcement Learning
Botunac, I., Petković, T., & Bosna, J. (2025) · Big Data and Cognitive Computing · doi:10.3390/bdcc9120317 · Cited by 1 · see citing works
“These stocks were selected to assess the system’s ability to interpret complex market signals and execute trading decisions effectively [73].”
Zhao, Y., Zhang, W., Yang, T., Jiang, Y., Huang, F., & Lim, W. Y. B. (2026) · ACM International Conference on Web Search and Data Mining (WSDM) · doi:10.1145/3773966.3777972 · Cited by 2 · see citing works
Self-citationFN-Agents: Analysis of Exchange Rate Volatility Prediction Based on Multi-Agent Systems
Guo, Z., Guo, W., Li, Q., et al. (2024) · International Conference on Computers and Artificial Intelligence Technology (CAIT) · doi:10.1109/cait64506.2024.10963060 · Cited by 1 · see citing works
“Although intelligent agents like FinAgent[7] and FinMem[8] have achieved some success in the financial sector, these technologies still face challenges, especially in the field of exchange rate forecasting, where real-time information retrieval and data processing rely on static data, making it difficult to reflect ma…”
QuantBench: benchmarking AI methods for quantitative investment from a full pipeline perspective
Wang, S., Kong, H., Guo, J., Hua, F., Qi, Y., Zhou, W., Zheng, J., Wang, X., Ni, L. M.-S., & Guo, J. (2025) · Frontiers of Information Technology & Electronic Engineering · doi:10.1631/fitee.2500280 · Cited by 1 · see citing works
Self-citationGeneralizable Multi-Agent Framework for Quantitative Trading of US Education Funds
Xu, K., Zhang, Z., Wilson, A., Yu, Q., Zhou, L., & Jiang, Y. (2024) · Innovations in Applied Engineering and Technology · doi:10.62836/iaet.v3i1.362
Multi-Modal AI Systems for Personalized Financial Planning
Selvam, M. (2025) · European Journal of Computer Science and Information Technology · doi:10.37745/ejcsit.2013/vol13n40187208
Graph LLM-Based Portfolio Management Algorithm
Qin, D., Yan, Y., & Kuruoğlu, E. E. (2025) · IEEE Conference on Artificial Intelligence (CAI) · doi:10.1109/cai64502.2025.00032
“LLM-based multimodal foundational agents are introduced and designed for financial trading, which processes textual, numerical, and visual data, incorporating expert insights and advanced reflection mechanisms to adapt to complex and volatile market dynamics [2].”
Unpacking the Decision Logic of LLM-based Agents: Evidence from the Newsvendor ProblemPreprint
Zhang, X., & Cavusoglu, H. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5337019
Large Language Models as Simulated Human Investors: A Paradox of Model SophisticationPreprint
Wang, J., Xu, Y., & Zhang, L. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5417134
Knowledge Discovery in Intelligent Business Systems：Current Research and Future Directions
Zhang, J., Wang, C., & Wang, L. (2025) · 中国科学基金 · doi:10.3724/bnsfc-2025-0014
PeterAI: An AI Multi-agent Framework for Personalized Investment Consultancy
Gadioli, A. V. A., Azevedo, P., Frizera, A., Souza, A. F. D., Oliveira-Santos, T., & Badue, C. S. (2025) · Communications in computer and information science · doi:10.1007/978-981-95-4091-4_9
Vision, Voice, and Text: Pioneering Zero-shot Multimodal LLMs for Sentiment-driven Investment
Tan, S., So, C. C., Sun, Y., Wang, J., Loh, A. W.-K., & Yung, S. P. (2025) · ACM International Conference on AI in Finance (ICAIF) · doi:10.1145/3768292.3770368
On the Potential of Tool-Enhanced Small Language Models to Match Large Models in Finance
Assis, G., Surica, A., Kroll, P., Munhoz, C., Rabbani, D., Bollis, E. R., Pellicer, L. F. A. O., & Paes, A. (2025) · International Conference on AI in Finance · doi:10.1145/3768292.3770409 · Cited by 1 · see citing works
Decision Alignment Protocols: Harmonising AI Agents for Comprehensive Market Assessment
Arabambi, A., Kayode, B. F., Faroun, O., et al. (2025) · International Conference on Electrical, Communication, and Computer Engineering (ICECCE) · doi:10.1109/icecce67514.2025.11257956
“Current multi-agent financial systems, including FinCon [5], TradingAgents [6], and FinAgent [7], demonstrate varying degrees of success in their respective domains but suffer from fundamental architectural limitations.”
InfluentialAn empirical study of self-admitted technical debt in AI agents
Fang, S., Gao, Z., Bi, T., et al. (2026) · Journal of Systems and Software · doi:10.1016/j.jss.2026.113068
Self-citationA Self-Improving Multi-Agent Framework for Dynamic Portfolio Management
Zhan, W., Yu, C., Xu, C., et al. (2025) · doi:10.1109/ica67499.2025.00025
“LLM-based agents can analyze diverse data sources, such as news, prices, reports and extract valuable insights [6].”
Agentic AI Meets Fundamentalist Financial Analysis: Limits and Possibilities
Nepomuceno, J., Rottschaefer, E., Assis, G., Ferreira, T. C., & Paes, A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-15987-8_9
Language Model Orchestrated Financial Agents: An Open-Source Framework
Gundimeda, R. T. (2025) · International Conference for Advancement in Technology (ICONAT) · doi:10.1109/iconat66879.2025.11362528
Improving LLM-Assisted Hydropower Operation Through Data Query Enhanced Agent
Luo, W., Zhang, F., Wang, L., Hu, S., Xu, B., & Zhao, J. (2026) · Lecture notes in civil engineering · doi:10.1007/978-981-95-4889-7_49
A Multimodal Agent for Stock Price Trend Forecasting
Zhu, X., Wang, X., & Du, H. (2025) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata66926.2025.11401488
“In financial scenarios, preliminary attempts have integrated textual and visual modalities and introduced external tool augmentation [11].”
Self-citationMulti Modal and Self Learning Agents in Finance
Yadav, A., Luqman, M., Desai, A. P., Gauba, H., & Yadav, P. (2025) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata66926.2025.11401653
“FINAGENT [17], discussed above, is a multimodal financial trading agent.”
He, Q., Zhu, X., Zhou, Y., Ding, F., & Chen, W. (2026) · International Conference on Intelligent Control and Information Processing · doi:10.1109/icicip67436.2026.11417616
“FinAgent proposed by Zhang et al. [35] demonstrated a multi-modal foundation agent that integrates numerical calculation tools, such as TA-Lib.”
Li, S., Lin, H., Liu, X., et al. (2026) · Artificial Intelligence · doi:10.1016/j.artint.2026.104539
FinLumen: A Game-Theoretic Multi-Agent Framework for Rational Portfolio Management
Zhang, Z., Sen, P., Gu, F., et al. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11462217
KG-ToolPlan: Knowledge Graph-Guided Reasoning for Efficient LLM Tool Selection
Cheng, Y., Wan, Z., Li, C., Wang, S., Ma, T., & Zhao, W. (2026) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp55912.2026.11461723
Event-Driven Financial Market Analyses Using a Multi-Agent Framework
Chen, Q., & Madisetti, V. K. (2026) · IEEE Transactions on Big Data · doi:10.1109/tbdata.2026.3688995
“Contemporary implementations of AI agents, exemplified by FinCon [8] for portfolio management, FinRobot [9] and FinAgent [10] for trading, and FinMem [11] for financial decision-making, reflect the increasing reliance on these technologies to execute complex financial tasks.”
Liu, J.-H., & Ma, Y. (2026) · International Conference on Advances in Artificial Intelligence and Machine Learning (AAIML) · doi:10.1109/aaiml67890.2026.11498104
“Understanding these failures helps identify biases (e.g. collective bullishness) and underscores the need for diverse or contrarian reasoning in the ensemble [10].”
FALLMAS: An LLM-Based Multi-Agent System for Automated Financial Analysis and Strategic Reporting
Bouchtib, H., Tabaa, M., Karboub, K., & Hamlich, M. (2026) · IEEE International Multidisciplinary Conference on Engineering Technology (IMCET) · doi:10.1109/imcet69180.2026.11503762
“FinAgent [22] and FinMem [23], require multi-modal data and layered memorization to outline inputs into memories and inform decisions.”
Multi-agent Reinforcement Learning for Financial Market Trading: An Expert-System Approach
Sevgili, S. O., & Öğüdücü, Ş. G. (2026) · Communications in computer and information science · doi:10.1007/978-3-032-19096-3_34
Complex Task Planning for Large Language Models via Text-Code Collaborative Reasoning
Tang, C., Long, Z., Wang, X., Wang, Z., & Li, G. (2026) · International Conference on Generative Artificial Intelligence and Information Security (GAIIS) · doi:10.1109/gaiis69281.2026.11519234
Self-citationA Review of Large Language Models for Stock Price Forecasting from a Hedge-Fund Perspective
Zhang, O., & Zhang, Z. (2026) · IEEE Conference on Artificial Intelligence (CAI) · doi:10.1109/cai68641.2026.11536649
“One advantage of multi-agent trading systems [59], [71], [60] is that they are well-rounded.”
The Rise of Generative AI in Finance: A Survey of Techniques and Studies
Feng, S., Tan, Y. L., Ren, J. S. J., Yeo, E. X. Y., Ong, R., Xie, M., & Wang, J. (2026) · Memetic Computing · doi:10.1007/s12293-026-00513-6
MultiHedge: Adaptive Coordination via Retrieval-Augmented Control
Bańka, F., & Chudziak, J. A. (2026) · Lecture notes in computer science · doi:10.1007/978-3-032-29924-6_39
Large language model-based multi-agent systems for financial markets simulation: a survey
Liu, Q., Yao, L., Wang, Z., Yang, Y., Tang, Y. J., Cheng, D., & Jiang, C. (2026) · Science China Information Sciences · doi:10.1007/s11432-026-4986-x · Cited by 2 · see citing works
Nguyen, P., & Pham, T. (2026) · Lecture notes in computer science · doi:10.1007/978-981-92-2014-4_26 · Cited by 2 · see citing works
175Chapter 8 Building Agentic AI Systems for Redefining Enterprise Applications
Manjula, A., Devi, S. C., Thomas, J. O., et al. (2026) · De Gruyter eBooks · doi:10.1515/9783112223062-008
Yue, S. L., Lu, J., Zhang, Z., et al. (2026) · IEEE Access · doi:10.1109/access.2026.3716870
“Agent-based approaches have emerged as a leading paradigm: FinA-gent [23] deploys multi-source LLM agents for autonomous trading decisions, and FinMem [24] introduces layered memory mechanisms for sequential financial reasoning over time.”
Multi-Domain Agentic Fraud Detection System
Putra, D. M. R., Parlambang, I. M. R., Manurung, P. N., et al. (2026) · IEEE International Conference on Industry 4.0, Artificial Intelligence, and Communications Technology (IAICT) · doi:10.1109/iaict71158.2026.11620884
“W. Zhang et al. [14] developed FinAgent, which combines market data analysis with trading tools to make investment decisions autonomously.”
Consensus-Gated Execution: A Multi-Agent LLM Architecture for Autonomous Cryptocurrency Trading
Bajgora, A., Kulakov, A., & Merovcı, F. (2026) · Electronics · doi:10.3390/electronics15153453
Causal-DFM: Diffusion-Based Causal-Invariant Dynamic Factor Model for Stock Return Prediction
Yin, Z., He, Y., Qin, X., Ren, J., Fu, S., & Wang, X. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3818050
Self-citationLifting the Veil of Non-Stationarity in Financial Market
Fu, V., Xu, X., Zhang, X., Xu, W., Ren, R., Deng, B., Zhao, X., Cao, J. W., & Cao, X. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3817881
Agents in the Wild: Where Research Meets Deployment
Yang, G. H., Venkit, P. N., Sedghamiz, H., Santus, E., Dibia, V. C., & Baldini, I. (2026) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/3770855.3816449
“For example, systems like FinAgent [55] and FinRobot [60] use multi-component architectures (LLM reasoning modules, financial data parsers, tool-using executors) to tackle tasks such as earnings call summarization, portfolio allocation, and risk assessment.”
Fin-DR: Learning DeepResearch Agent for Financial Time-Series ForecastingPreprint
Duan, J., Zhang, Y., Junfu, Z., et al. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7422697
TextGrad-Trader: Risk-Aware Multimodal Signal Arbitration for Adaptive Stock Prediction ServicesPreprint
Wu, T., Gong, W., Wang, D., et al. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7455222
Self-citationA Causality-Gated Multi-Agent System for Generative Feature Engineering in Financial MarketsPreprint
Karaman, M. E., & Üstündağ, A. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.7493407
A Resource-Aware Selectively Agentic Framework for A-Share Trading
Chen, X., Jin, X., & Zhang, D. (2026) · IEEE International Conference on Pattern Recognition and Machine Learning (PRML) · doi:10.1109/prml71097.2026.11687303
“Recent work therefore organizes financial reasoning into analyst, trader, memory, and risk roles, arguing that division of labor is a more realistic basis for machine-supported investment decisions than monolithic forecasting alone [1]-[5].”
GroundTrader: Multi-Agent Transition Supervision for LLM-Based Financial Trading
Cui, X., Liu, Y., & Zhang, S. (2026) · Mathematics · doi:10.3390/math14193472
RealMem: Benchmarking LLMs in Real-World Memory-Driven Interaction
Bian, H., Yao, Z., Hu, S., Xu, Z., Zhang, S., Guo, Y., Yang, Z., Han, X., Wang, H., & Chen, R. (2026) · arXiv · doi:10.48550/arxiv.2601.06966
“…is essential not only for practical applications such as personalized chatbots (Li et al., 2025a) and financial analysis (Zhang et al., 2024), but also as a key enabler toward Artificial General Effective long-term memory is essential for agents to maintain coherent collaboration across…”
- Authors
- Zhang, W., Zhao, L., Xia, H., Sun, S., Sun, J., Qin, M., Li, X., Zhao, Y., Zhao, Y., Cai, X., Zheng, L., Wang, X., & An, B.
- Affiliations
- Nanyang Technological University (SG)
- National Technological University, Singapore, Singapore
- National University of Singapore (SG)
- ORCID
- Venue
- Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining
- Publisher
- ACM
- Published
- 2024-08-24
- Pages
- 4314-4325
- OpenAlex
- W4401863645
- Cited by
- 74
- References
- 25
- Citations by year
- 2026: 32 · 2025: 38 · 2024: 4
- FWCI
- 50.60
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) (publishedVersion)
- DR-NTU (Nanyang Technological University) (publishedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- arXiv (Cornell University) (submittedVersion, cc-by)
- Singapore Management University Institutional Knowledge (InK) (Singapore Management University) (submittedVersion, cc-by-nc-nd)
- arXiv (Cornell University) (cc-by)
- Funders
- National Research Foundation
- National Research Foundation Singapore
- 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 › Accounting › Financial Analysis and Corporate Governance
- Keywords
- financial trading, market intelligence, memory retrieval, trading decision-making, high-frequency trading, quantitative trading, business news
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.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.
Cited sources are checked by 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
