Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

ElliottAgents: A Natural Language-Driven Multi-Agent System for Stock Market Analysis and Prediction

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper reports that a natural-language multi-agent system with LLM agents can detect Elliott Wave patterns, backtest them with reinforcement learning, and issue concrete buy/sell recommendations with Fibonacci price targets.

desk verdict A well-written system paper whose effectiveness claim is unsupported by its own circular and thin evaluation. read the letter →

arxiv 2507.03435 v1 pith:QHTIAPNF submitted 2025-07-04 cs.CE

Jaros{l}aw A. Chudziak , Micha{l} Wawer This is my paper
classification cs.CE
keywords multi-agentsystemslargelanguagemodelsElliottWavePrinciplestockmarketpredictionnaturalprocessingdeepreinforcementlearningretrieval-augmentedgenerationfinancialNLP
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper sets out to show that an ensemble of LLM-powered agents, asked to converse in natural language, can perform Elliott Wave analysis on historical stock data and convert the detected patterns into concrete, readable trading plans. The motivation is that technical analysis like the Elliott Wave Principle is pattern-based and language-heavy, so a conversational multi-agent design should do better than black-box predictors at both accuracy and interpretability. If the claim holds, investors would get transparent, explainable price forecasts with explicit targets and dates, and NLP-based agent systems would have a demonstrated use case in data-intensive finance. The authors report case-study profits of 4.4%, 13.3% and 17.4% on AMZN, GOOG and NVDA, plus cross-validation accuracies in which complete impulsive waves outperform incomplete ones and DRL backtesting improves prediction.

What carries the argument

The load-bearing machinery is the Elliott Wave Principle (EWP) embedded in a multi-agent LLM pipeline. EWP holds that prices move in five-wave impulsive patterns and three-wave corrective patterns whose lengths and retracements obey Fibonacci ratios, and that these patterns are fractal across timeframes. ElliottAgents adds a custom tool that finds all possible impulsive and corrective wave patterns, a DRL-based backtester that checks whether patterns worked on historical data, a RAG module that feeds agents the mathematical theory of EWP, and a hierarchical dialogue in which the Data Engineer, Elliott Waves Analyst, Backtester, Technical Analysis Expert, Investment Advisor and Reports Writer refine one another's conclusions. The wave-detection tool's labels are simultaneously the trading signal and the success criterion, so the EWP pattern definitions and the Fibonacci rules for target prices carry the argument.

What would settle it

Run the same cross-validation protocol on the same stocks and intervals but replace the wave-detection tool's labels with randomly placed patterns; if the reported prediction accuracies and theoretical returns survive intact, the Elliott Wave labels are not doing the predictive work.

Watch

Extended reading notes

Core claim

On its own terms, the central discovery is that a hierarchy of LLM agents can operationalize the Elliott Wave Principle: a custom tool enumerates all possible impulsive and corrective wave patterns, a DRL backtester filters those patterns by historical reliability, and the agents' natural-language dialogue lets a Technical Analysis Expert and Investment Advisor turn the surviving patterns into buy/sell decisions with Fibonacci-derived price targets. The system's predictions are considered correct when the average price over the next $n$ candlesticks moves in the signaled direction, where $n$ follows EWP wave-length rules. Using that criterion, complete five-wave impulse patterns gave higher accuracies than incomplete four-wave patterns, and adding DRL backtesting improved accuracy in nearly every stock-interval combination. The authors therefore claim the agents' collaboration is effective enough for simulated transactions to yield the theoretical returns reported in the case studies.

Load-bearing premise

The load-bearing premise is that the in-house tool's wave labels are correct and that Elliott Wave patterns genuinely contain predictive information; the paper never validates the labeling against human experts or random labels, so if the labels are arbitrary, the reported accuracies and profits follow from the labeling choice rather than from real market structure.

Editorial extensions

If this is right

  • Complete impulsive five-wave patterns are more reliable predictors of the next leg than incomplete four-wave patterns; this is a testable ranking within EWP.
  • DRL backtesting improves directional forecast accuracy for most stock-interval pairs, suggesting reinforcement learning can act as a filter over technical patterns.
  • The agent pipeline yields explicit price targets, dates, and buy/sell signals derived from Fibonacci wave relationships, so the output is actionable by a lay trader.
  • Natural-language dialogue among specialized agents is a viable decomposition strategy for complex financial analysis tasks.
  • The success criterion based on average price of the next $n$ candlesticks provides a simple, reproducible evaluation protocol for wave-based predictions.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Not claimed by the paper: the evaluation uses the wave tool's own labels as both signal and ground truth, so an independent check against human Elliott Wave labeling or a random-label baseline is the unstated next experiment.
  • Not claimed by the paper: the case-study gains of 4.4%, 13.3% and 17.4% are theoretical price moves and do not subtract transaction costs or slippage; realistic paper trading could change them materially.
  • Not claimed by the paper: the hierarchical dialogue pattern could be lifted into other rule-intensive interpretive fields, such as candlestick pattern analysis, but the paper does not test that transfer.
  • Not claimed by the paper: because completed impulses outperform incomplete ones, preferring completed patterns and requiring confirmation might improve future versions; the paper only reports the comparison.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper presents ElliottAgents, a multi-agent system in which LLM-based agents collaborate through natural-language dialogue to perform stock-market analysis and prediction guided by the Elliott Wave Principle (EWP). The architecture includes a data engineer, an Elliott Waves Analyst that uses an in-house tool to detect wave patterns, a DRL-based backtester, a technical analysis expert, an investment advisor, and a reports writer. The evaluation has two parts: three narrative case studies on AMZN, GOOG, and NVDA with claimed theoretical profits, and a cross-validation experiment on 1000 candlesticks for six stock/interval combinations, reporting directional accuracy for incomplete (1-2-3-4) and complete (1-2-3-4-5) impulsive patterns with and without DRL backtesting. The paper concludes that these results demonstrate the effectiveness of the agent collaboration and its usefulness for real trading.

Significance. If the effectiveness claims were substantiated, the system would be a noteworthy applied contribution: it combines LLM-based multi-agent coordination with a classical technical-analysis framework, and the agent dialogue/reporting component addresses a real interpretability need in financial NLP. The strengths are the concrete agent decomposition, the use of RAG to ground outputs in EWP theory, and the attempt to combine symbolic pattern detection with DRL. However, the current evidence does not support the central claim: the wave detector is unvalidated, evaluation horizons are derived from the same detector, sample sizes are tiny, no baselines or uncertainty estimates are provided, and the reported profits come from selected successful cases. The application-level idea is plausible, but the paper does not yet demonstrate it.

major comments (5)
  1. [§4.2, Table 1] The quantitative evaluation in Table 1 cannot support the accuracy claims because the per-cell pattern counts are extremely small (N=8 to 28) and no confidence intervals, significance tests, or baselines are reported. For example, with N=8, the 62.50% accuracy in the AMZN hourly 1-2-3-4-5 column corresponds to 5/8 correct, and the 95% binomial interval spans roughly 24% to 91%, so the differences between conditions are within sampling noise. The paper also does not describe the cross-validation split (e.g., how the 1000 candlesticks are partitioned, whether adjacent samples overlap, or how the DRL training window relates to the test window), preventing an assessment of leakage or overfitting.
  2. [§3.1 and §4.2] The central signal is generated by an in-house tool whose output is never validated. Section 3.1 states that the authors "create a special tool that finds all possible impulsive and corrective wave patterns in the data," but no evidence is given that these automatically detected waves correspond to what a trained Elliott Wave analyst would label, nor is the detector compared with a null model on random or permuted price series. This matters because the same detector's labels define both the predicted direction and the prediction horizon n (the length of wave 5, or 1.62 times wave 1, as described in the evaluation paragraph of §4.2). A permissive detector can always find a wave-like structure, and a horizon constructed from the detector's own geometry can make a 'correct' direction call appear more likely than it is. At minimum, the detector should be validated against human annotations and the evaluation should use fixed, pre-registered horizons and independent labels.
  3. [§4.2, Figs. 6–8] The three narrative case studies are selected successful trades and do not constitute a test of the system. The text reports profits of 4.4%, 13.3%, and 17.4% on AMZN, GOOG, and NVDA, but it does not report how many signals were issued in the same period, how many failed, what the holding periods were, what transaction costs and slippage would be, or what a buy-and-hold or index benchmark returned over the same intervals. The claim that simulated transactions "prove the effectiveness of our agent's collaboration" (end of the second paragraph of §4.2) is therefore unsupported. A complete signal log with all trades, including losses, and a comparison against a benchmark is required.
  4. [§4.2 and §3.2] The comparison "with backtesting" versus "without backtesting" is not a controlled experiment. Section 3.2 describes the DRL backtester as learning on historical data, and §4.2 says it was run on 10 years of historical data for each company, but the paper does not state whether the DRL policy was trained on the same 1000-candlestick evaluation window, how the training and test periods were separated under the claimed cross-validation, which DRL algorithm and reward function were used, or how hyperparameters were selected. Without this information, the observed improvements (e.g., 53.57% to 67.86% for GOOG daily 1-2-3-4 patterns) could result from in-sample fitting rather than from the backtester adding useful out-of-sample information.
  5. [§3.2 and §4.2] The paper does not identify the LLM(s) used by the agents, their version, prompt templates, temperature, or any of the wave-detection thresholds, making the experiments unrepeatable. Since the authors mention GPT-4 as an example but never state which model generated the analyses in Figs. 6–8 and Table 1, it is impossible to separate the contribution of the multi-agent architecture from the choice of a particular LLM. Reporting these configuration details is necessary for any claim of demonstrated effectiveness.
minor comments (5)
  1. [Introduction and §3.1] The text contains grammatical errors such as "an multi-agent system," "a established method," and "Eliott waves analyst"; these should be corrected.
  2. [§2.3] The description of large language models is repeated almost verbatim in two consecutive paragraphs; the redundant passage should be removed.
  3. [§4.2] The phrase "resulted in a improvement" is ungrammatical and should read "resulted in an improvement."
  4. [§5.1] The claim of "better accuracy than similar systems using EWP (Tirea et al., 2012)" is made without reporting the comparison system's numbers or reproducing its experimental setting; either provide the direct comparison or soften the claim.
  5. [§3.3] The relationship between the agent flow diagram in Fig. 5 and the architecture diagram in Fig. 3 is not explained; the text should clarify how the use-case workflow maps onto the system architecture.

Circularity Check

2 steps flagged · score 6.0 of 10

ElliottAgents' reported prediction accuracy and chart targets are scored using horizons and price levels derived from the same unvalidated Elliott-wave labels that generate the signals, so the central effectiveness claim is partially circular.

  1. self definitional [Section 4.2, 'Evaluation of Use Cases' (penultimate paragraph)]
    "Based on the identified patterns, agents predicted whether the next movement would be upward or downward. A prediction was considered correct if the average price of the subsequent n candlesticks was higher or lower, depending on the issued signal. The n number of candlesticks was determined according to EWP, where in the case of waves 1-2-3-4, the length of the fifth wave should be approximately 1.62 times the length of the first wave, and in the case of a complete impulse wave, the following wave A should have a length close to wave 5."

    The predicted direction and the evaluation horizon are both read off the same detected Elliott wave pattern that issued the signal. A detected 1-2-3-4 pattern is by the system's rule set supposed to be followed by an upward fifth wave, and a detected complete 1-2-3-4-5 is supposed to be followed by a downward A wave; the horizon n is then set to the detected wave's own scaled length (1.62 times wave 1, or the length of wave 5). A 'correct' call therefore does not test predictive power: it checks whether prices moved within a horizon constructed from the detector's own labels. Any sufficiently permissive wave finder can manufacture correct calls by labeling waves and rescaling n accordingly.

  2. fitted input called prediction [Section 4.2, Fig. 7 (Alphabet daily interval)]
    "The target price was strategically set at $160 per share, aligning with the peak of the fifth wave, while also accounting for the resistance level observed at the peak of wave B ($150). ... As the experimental data, in the chart shows, price levels have been achieved."

    The 'target' is not an independent forecast: it is defined as the peak of the fifth wave as labeled by the same unvalidated wave-detection tool that generated the trade signal. When the paper later reports that 'price levels have been achieved', it is comparing the chart against the tool's own post-hoc wave annotations. This is equivalent to fitting a curve through data points and then 'predicting' those same points; it does not demonstrate that ElliottAgents can forecast future target prices.

full rationale

The central empirical claim—'Based on these signals, we could simulate transactions and calculate theoretical investment returns, proving the effectiveness of our agent's collaboration'—rests on two self-referential constructions. First, the correctness of a direction call is judged over n candlesticks where n is set by the detected Elliott wave geometry, and the direction itself is the one dictated by the detected pattern; thus the 'prediction' is a restatement of the labeling rule, and the evaluation only confirms that prices moved within a horizon the tool constructed from its own labels. Second, chart targets are set at the peaks of waves identified by the same unvalidated tool; reporting that price levels were reached is then comparing the chart to the tool's own post-hoc annotations, not to an independent forecast. The evaluation lacks a null model or human-annotation check for the wave detector, so this circularity is not resolved. The DRL comparison is additionally vulnerable to in-sample fitting because it uses 10 years of the same companies' data with no described temporal split, but that concern is secondary. Because the pattern-detection architecture itself may contain useful content beyond these scoring choices, the circularity is partial rather than total: score 6.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The central evaluation rests on assumptions pulled from Elliott Wave theory and from the authors' own unvalidated pattern detector; none of these are justified with independent evidence in the paper. No new physical or conceptual entities are postulated; the six named agents are implementation modules rather than scientific constructs.

free parameters (2)
  • DRL backtesting policy parameters = undisclosed
    The Backtester trains on 10 years of historical data for each stock (Section 4.2); the policy and reward weights are fitted and not reported.
  • Wave detection thresholds = undisclosed
    The special tool that finds all possible impulsive and corrective wave patterns (Section 3.1) must have tolerance parameters for Fibonacci ratios and overlap constraints; none are given.
assumptions (5)
  • domain assumption Elliott Wave Principle is a valid predictive framework for stock prices
    Invoked in Section 2.2 and used to define all signals, target prices, and prediction horizons; its predictive validity is not tested.
  • ad hoc to paper The in-house tool correctly identifies Elliott wave patterns
    Section 3.1 states 'we create a special tool that finds all possible impulsive and corrective wave patterns'; no ground truth or inter-annotator agreement is provided.
  • domain assumption Prediction horizon n follows EWP length ratios (fifth wave ~1.62 times first wave, wave A close to wave 5)
    Section 4.2 uses this rule to set the number of candlesticks over which accuracy is measured; the rule comes from Elliott theory, not from data.
  • domain assumption Average price direction over the next n candlesticks is a meaningful measure of prediction quality
    Section 4.2 defines correctness this way; no justification that this metric relates to tradable returns or risk.
  • domain assumption DRL backtesting on 10 years of the same company's history improves out-of-sample interpretation
    Section 4.2 attributes accuracy improvements to the backtester, but the training and evaluation periods may overlap.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ElliottAgents: A Natural Language-Driven Multi-Agent System for Stock Market Analysis and Prediction." pith.science (2026). https://pith.science/paper/QHTIAPNF

@misc{pith2026250703435,
  author       = {Pith},
  title        = {Pith review of: ElliottAgents: A Natural Language-Driven Multi-Agent System for Stock Market Analysis and Prediction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QHTIAPNF}},
  note         = {Machine review of arXiv:2507.03435}
}
read the original abstract

This paper presents ElliottAgents, a multi-agent system leveraging natural language processing (NLP) and large language models (LLMs) to analyze complex stock market data. The system combines AI-driven analysis with the Elliott Wave Principle to generate human-comprehensible predictions and explanations. A key feature is the natural language dialogue between agents, enabling collaborative analysis refinement. The LLM-enhanced architecture facilitates advanced language understanding, reasoning, and autonomous decision-making. Experiments demonstrate the system's effectiveness in pattern recognition and generating natural language descriptions of market trends. ElliottAgents contributes to NLP applications in specialized domains, showcasing how AI-driven dialogue systems can enhance collaborative analysis in data-intensive fields. This research bridges the gap between complex financial data and human understanding, addressing the need for interpretable and adaptive prediction systems in finance.

Figures

Figures reproduced from arXiv: 2507.03435 by the authors.

Figure 2
Figure 2. Fibonacci retracements in corrective waves [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Data flow between agents. understanding, reasoning, and decision-making ca￾pabilities to agents. LLMs enable agents to op￾erate more autonomously, adapting to new situa￾tions without requiring explicit instructions. These agents can now exhibit goal-directed behaviors, making proactive decisions to achieve long-term objectives, enhancing their autonomy and proac￾tiveness (Guo et al., 2024; Cinkusz and Chudziak, 2024… view at source ↗
Figure 4
Figure 4. Overview of a LLM autonomous agent, based [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (3 more)
Figure 6
Figure 6. Figure 6: Ending diagonal pattern recognized on AMZN [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 7
Figure 7. Figure 7: Fifth wave extension found on GOOG 1d chart. profit realized from this transaction amounted to $20 per share, translating to a 13.3% gain. In the [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 8
Figure 8. Figure 8: Full wave cycle recognized on NVDA 1d [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Securing Multi-Agent Systems Against Corruptions via Node Contribution Backpropagation

    cs.CR 2025-10 conditional novelty 4.0 of 10

    A backward-propagation scoring scheme over a signed temporal DAG can identify malicious agents in LLM multi-agent systems and cut their communications, improving defended accuracy by 3–7 percentage points in the autho...

Reference graph

Works this paper leans on

37 extracted references · 25 canonical work pages · cited by 1 Pith paper

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Akintola and O.E

    K.G. Akintola and O.E. Oyetunji. 2021. Development of an agent-based framework for stock market trading. IRE Journals, 4(9)

    work page 2021

  4. [4]

    Thimira Amaratunga. 2023. Understanding Large Language Models Learning Their Underlying Concepts and Technologies. apress

    work page 2023

  5. [5]

    Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2023. Self-rag: Learning to retrieve, generate, and critique through self-reflection. arXiv:2310.11511 [cs.CL]

    arXiv 2023

  6. [6]

    Carolyn Boroden. 2008. Fibonacci Trading: How to Master the Time and Price Advantage. McGraw Hill

    work page 2008

  7. [7]

    Haiping Chen. 2024. Understand the llm agent orchestration. https://medium.com/scisharp/understand-the-llm-agent-orchestration-043ebfaead1f. Accessed: Jun. 1, 2024

    work page 2024

  8. [8]

    Adam Chudziak. 2023. Predictability of stock returns using neural networks: Elusive in the long term. Expert Systems with Applications, 213

    work page 2023

Show all 37 references
  1. [9]

    Chudziak and Konrad Cinkusz

    Jaroslaw A. Chudziak and Konrad Cinkusz. 2024. Towards llm-augmented multiagent systems for agile software engineering. In The 39th IEEE/ACM International Conference on Automated Software Engineering (ASE 2024), Sacramento, CA, USA

    2024

  2. [10]

    Chudziak

    Konrad Cinkusz and Jaroslaw A. Chudziak. 2024. Communicative agents for software project management and system development. In Proceedings of the 21th International Conference on Modeling Decisions for Artificial Intelligence (MDAI 2024), Tokyo, Japan

    2024

  3. [11]

    A. J. Frost, Robert R. Prechter Jr., and Charles J. Collins. 2001. Elliott Wave Principle: Key to Market Behavior. Wiley

    2001

  4. [12]

    Gamil, Raafat S

    Ahmed A. Gamil, Raafat S. El-fouly, and Nevin M. Darwish. 2007. Stock technical analysis using multi agent and fuzzy logic. In Proceedings of the World Congress on Engineering, WCE 2007, London, UK

    2007

  5. [13]

    Taicheng Guo et al. 2024. Large language model based multi-agents: A survey of progress and challenges. arXiv:2402.01680v2 [cs.CL]

    2024 arXiv

  6. [14]

    Taylan Kabbani and Ekrem Duman. 2022. Deep reinforcement learning approach for trading automation in the stock market. IEEE Access, 10

    2022

  7. [15]

    Hobson Lane, Hannes Hapke, and Cole Howard. 2019. Natural Language Processing in Action: Understanding, analyzing, and generating text with Python. Manning

    2019

  8. [16]

    Maxim Lapan. 2020. Deep Reinforcement Learning Hands-On Second Edition. Packt

    2020

  9. [17]

    Jonathan Larson and Steven Truitt. 2024. Graphrag: Unlocking llm discovery on narrative private data. https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data. Accessed: May. 10, 2024

    2024

  10. [18]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2021. Retrieval-augmented generation for knowledge-intensive nlp tasks. arXiv:2005.1140...

    2021 arXiv

  11. [19]

    Junyou Li, Qin Zhang, Yangbin Yu, Qiang Fu, and Deheng Ye. 2024. More agents is all you need. arXiv:2402.05120v1 [cs.CL]

    2024 arXiv

  12. [20]

    Louie Peters Louis-François Bouchard. 2024. Building LLMs for Production: Enhancing LLM Abilities and Reliability with Prompting, Fine-Tuning, and RAG. Towards AI

    2024

  13. [21]

    Luo, Kecheng Liu, and Darryl N

    Y. Luo, Kecheng Liu, and Darryl N. Davis. 2002. A multi-agent decision support system for stock trading. IEEE Network, 16(1)

    2002

  14. [22]

    Johann Lussange, Ivan Lazarevich, Sacha Bourgeois-Gironde, Stefano Palminteri, and Boris Gutkin. 2020. https://doi.org/10.1007/s10614-020-10038-w Modelling stock markets by multi-agent reinforcement learning . Computational Economics. Hal-03055070

    2020 doi

  15. [23]

    John J. Murphy. 1999. Technical Analysis of the Financial Markets: A Comprehensive Guide to Trading Methods and Applications. New York Institute of Finance

    1999

  16. [24]

    Humza Naveed, Asad Ullah Khan, Shi Qiu, Muhammad Saqib, Muhammad Usman Saeed Anwar, Naveed Akhtar, Nick Barnes, and Ajmal Mian. 2024. A comprehensive overview of large language models. arXiv:2307.06435 [cs.CL]

    2024 arXiv

  17. [25]

    OpenAI. 2023. Gpt-4 technical report

    2023

  18. [26]

    Saddam Hossain Mukta, Kaniz Fatema, Nur Mohammad Fahad, Sadman Sakib, Most Marufatul Jannat Mim, Jubaer Ahmad, Mohammed Eunus Ali, and Sami Azam

    Mohaimenul Azam Khan Raiaan, Md. Saddam Hossain Mukta, Kaniz Fatema, Nur Mohammad Fahad, Sadman Sakib, Most Marufatul Jannat Mim, Jubaer Ahmad, Mohammed Eunus Ali, and Sami Azam. 2024. A review on large language models: Architectures, applications, taxonomies, open issues and ...

    2024

  19. [27]

    Szydlowski and Jaroslaw A

    Kamil L. Szydlowski and Jaroslaw A. Chudziak. 2024 a . Toward predictive stock trading with hidformer integrated into reinforcement learning strategy. In The 36th International Conference on Tools for Artificial Intelligence (ICTAI 2024), Herndon, VA, USA

    2024

  20. [28]

    Szydlowski and Jaroslaw A

    Kamil L. Szydlowski and Jaroslaw A. Chudziak. 2024 b . Transformer-style neural network in stock price forecasting. In The 21th International Conference on Modeling Decisions for Artificial Intelligence (MDAI 2024), Tokyo, Japan

    2024

  21. [29]

    Merrill, Vinayak Gupta, Tim Althoff, and Thomas Hartvigsen

    Mingtian Tan, Mike A. Merrill, Vinayak Gupta, Tim Althoff, and Thomas Hartvigsen. 2024. Are language models actually useful for time series forecasting? arXiv:2406.16964v1 [cs.LG]

    2024 arXiv

  22. [30]

    Hua Tang, Chong Zhang, Mingyu Jin, Qinkai Yu, Zhenting Wang, Xiaobo Jin, Yongfeng Zhang, and Mengnan Du. 2024. Time series forecasting with llms: Understanding and enhancing model capabilities. arXiv:2402.10835v2 [cs.CL]

    2024 arXiv

  23. [31]

    Monica Tirea, Ioan Tandau, and Viorel Negru. 2012. Stock market multi-agent recommendation system based on the elliott wave principle. In International Conference on Availability, Reliability, and Security, Prague, Czech Republic

    2012

  24. [32]

    Lewis Tunstall, Leandro von Werra, and Thomas Wolf. 2022. Natural Language Processing with Transformers, Revised Edition. O'Reilly

    2022

  25. [33]

    Tomas Vantuch, Ivan Zelinka, and Pandian Vasant. 2016. Market prices trend forecasting supported by elliott wave’s theory. In First EAI International Conference on Computer Science and Engineering, Penang, Malaysia

    2016

  26. [34]

    Lilian Weng. 2024. Llm powered autonomous agents. https://lilianweng.github.io/posts/2023-06-23-agent/. Accessed: Jun. 1, 2024

    2024

  27. [35]

    Frank Wittkampf. 2024. Next-level agents: Unlocking the power of dynamic context. https://towardsdatascience.com/next-level-agents-unlocking-the-power-of-dynamic-context-68b8647eef89. Accessed: Jun. 1, 2024

    2024

  28. [36]

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. 2023. React: Synergizing reasoning and acting in language models. arXiv:2210.03629 [cs.CL]

    2023 arXiv

  29. [37]

    Pengyu Zhao, Zijian Jin, and Ning Cheng. 2023. An in-depth survey of large language model-based artificial intelligence agents. arXiv:2309.14365v1 [cs.CL]

    2023 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.