Paper: arXiv 2402.18485

Abstract

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 9 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.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — practical and empirical, lighter on theory

Why this score: The paper focuses on system architecture and multimodal data integration with advanced AI (LLMs, RL, vision), but lacks dense mathematical derivations or novel proofs; it is highly empirical, with specific backtest results, multiple datasets, and clear implementation details for a deployable agent.

Research Flowchart

  flowchart TD
  Start["Research Goal: Create Multimodal Foundation Agent for Financial Trading"]
  Data["Market Data: News, Prices, Kline Charts (Stocks & Crypto)"]
  Method["Methodology: FinAgent Architecture"]
  Process["Computational Processes: Market Intelligence & Dual-Level Reflection"]
  Tool["Tool-Augmented Strategy Integration"]
  Out["Outcome: 36% Avg Profit Improvement, 92.27% Return on Specific Dataset"]
  Start --> Data
  Data --> Method
  Method --> Process
  Process --> Tool
  Tool --> Out

Cited References (140)

  1. Soft actor-critic algorithms and applications (2018) — link
  2. Training verifiers to solve math word problems (2021) — link
  3. Webgpt: Browser-assisted question-answering with human feedback (2021) — link
  4. Lamda: Language models for dialog applications (2022) — link
  5. Llama: Open and efficient foundation language models (2023) — link
  6. Proximal policy optimization algorithms (2017) — link
  7. Playing atari with deep reinforcement learning (2013) — link
  8. FinGPT: Open-Source Financial Large Language Models — → 2306.06031
and 132 more
  1. FinMem: A performance-enhanced LLM trading agent with layered memory and character design (2023) — link
  2. Efficient capital markets: A review of theory and empirical work (1970)
  3. Stock market volatility and learning (2016)
  4. Prune and tune ensembles: low-cost ensemble learning with sparse independent subnetworks (2022)
  5. No One Representation to Rule Them All: Overlapping Features of Training Methods (2022)
  6. An adaptive master-slave regularized model for unexpected revenue prediction enhanced with alternative data (2020)
  7. Computational intelligence and financial markets: A survey and future directions (2016)
  8. Uncertainty-aware lookahead factor models for quantitative investing (2020)
  9. Listening to chaotic whispers: A deep learning framework for news-oriented stock trend prediction (2018)
  10. The theory of the trading firm revisited (1978)
  11. Stock movement prediction from tweets and historical prices (2018)
  12. Investment behaviors can tell what inside: Exploring stock intrinsic properties for stock trend prediction (2019)
  13. Deep ensembles: A loss landscape perspective (2019) — link
  14. Stock selection via spatiotemporal hypergraph attention network: A learning to rank approach (2021)
  15. Stock market’s price movement prediction with LSTM neural networks (2017)
  16. CLVSA: A convolutional LSTM based variational sequence-to-sequence model with attention for predicting trends of financial markets (2019)
  17. Enhancing stock movement prediction with adversarial training (2018) — link
  18. Hierarchical multi-scale Gaussian transformer for stock movement prediction. (2020)
  19. Temporal relational ranking for stock prediction (2019)
  20. Deep attentive learning for stock movement prediction from social media text and company correlations (2020)
  21. Qlib: An AI-oriented quantitative investment platform (2020) — link
  22. Accurate multivariate stock movement prediction via data-axis transformer with multi-level contexts (2021)
  23. Learning multiple stock trading patterns with temporal routing adaptor and optimal transport (2021)
  24. The Sharpe ratio (1994)
  25. When quantitative trading meets machine learning: A pilot survey (2016)
  26. Algorithmic trading (2011)
  27. A comprehensive evaluation of ensemble learning for stock-market prediction (2020)
  28. Stock price prediction via discovering multi-frequency trading patterns (2017)
  29. Rest: Relational event-driven stock trend forecasting (2021)
  30. Long-tailed recognition by routing diverse distribution-aware experts (2020)
  31. Stock investment decision support using an ensemble of L-GEM based on RBFNN diverse trained from different years (2012)
  32. Predicting the stock market using multiple models (2006)
  33. Hybrid forecasting model research on stock data mining (2010)
  34. Bagging predictors (1996)
  35. DoubleEnsemble: A new ensemble method based on sample reweighting and feature selection for financial data analysis (2020)
  36. Random classification noise defeats all convex potential boosters (2010)
  37. Data arameters: A new family of parameters for learning a differentiable curriculum (2019)
  38. CatBoost: Unbiased boosting with categorical features (2018)
  39. Stock market forecast using multivariate analysis with bidirectional and stacked (LSTM, GRU) (2018)
  40. Multi-scale two-way deep neural network for stock trend prediction. (2020)
  41. Alphastock: A buying-winners-and-selling-losers investment strategy using interpretable deep reinforcement attention networks (2019)
  42. VolTAGE: Volatility forecasting via text-audio fusion with graph convolution networks for earnings calls (2020)
  43. Adaptive mixtures of local experts (1991)
  44. Learning factored representations in a deep mixture of experts (2013) — link
  45. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer (2017) — link
  46. Modeling task relationships in multi-task learning with multi-gate mixture-of-experts (2018)
  47. Scaling vision with sparse mixture of experts (2021)
  48. Deep mixture of experts via shallow embedding (2020)
  49. GLaM: Efficient scaling of language models with mixture-of-experts (2021) — link
  50. A dual-stage attention-based recurrent neural network for time series prediction (2017)
  51. Designing financial strategies based on artificial neural networks ensembles for stock markets (2018)
  52. Stock market index prediction using deep neural network ensemble (2017)
  53. Evaluating multiple classifiers for stock price direction prediction (2015)
  54. Performance of moving average trading strategies over varying stock market conditions: The Finnish evidence (2014)
  55. Deep learning with gated recurrent unit networks for financial sequence predictions (2018)
  56. A LSTM-based method for stock returns prediction: A case study of China stock market (2015)
  57. Stock market value prediction using neural networks (2010)
  58. Coherent measures of risk (1999)
  59. Conditional value-at-risk for general loss distributions (2002)
  60. Measuring systemic risk (2017)
  61. Measuring uncertainty (2015)
  62. Modeling financial uncertainty with multivariate temporal entropy-based curriculums (2021)
  63. Uncertainty in deep learning (2016)
  64. Selective Ensembles for Consistent Predictions (2021)
  65. BatchEnsemble: an Alternative Approach to Efficient Ensemble and Lifelong Learning (2019)
  66. Hyperparameter ensembles for robustness and uncertainty quantification (2020)
  67. Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time (2022) — link
  68. Film: Visual reasoning with a general conditioning layer (2018)
  69. Flipout: Efficient Pseudo-Independent Weight Perturbations on Mini-Batches (2018)
  70. Simple and scalable predictive uncertainty estimation using deep ensembles (2017)
  71. Snapshot ensembles: Train 1, get m for free (2017) — link
  72. Loss surfaces, mode connectivity, and fast ensembling of dnns (2018)
  73. Neural network ensembles (1990)
  74. Scalable bayesian optimization using deep neural networks (2015)
  75. A survey of uncertainty in deep neural networks (2021) — link
  76. Neural network ensembles, cross validation, and active learning (1994)
  77. Efficient and robust automated machine learning (2015)
  78. Hyperparameter optimization (2019)
  79. Popular ensemble methods: An empirical study (1999)
  80. Deep learning (2015)
  81. Ensemble selection from libraries of models (2004)
  82. What is being transferred in transfer learning? (2020)
  83. Averaging weights leads to wider optima and better generalization (2018)
  84. Cognitive architectures for language agents (2023) — link
  85. Generative Agents: Interactive simulacra of human behavior (2023) — link
  86. Toolformer: Language models can teach themselves to use tools (2023) — link
  87. Palm: Scaling language modeling with pathways (2023)
  88. Bloomberggpt: A large language model for finance (2023) — link
  89. Voyager: An open-ended embodied agent with large language models (2023)
  90. PRUDEX -Compass: Towards systematic evaluation of reinforcement learning in financial markets (2023)
  91. Reinforcement learning for quantitative trading (2023)
  92. TradeMaster: A holistic quantitative trading platform empowered by reinforcement learning (2023)
  93. Mastering stock markets with efficient mixture of diversified trading experts (2023)
  94. Random search for hyper-parameter optimization. (2012)
  95. Quantitative trading: how to build your own algorithmic trading business (2021)
  96. Deep reinforcement learning for quantitative trading: Challenges and opportunities (2022)
  97. FinRL : A deep reinforcement learning library for automated stock trading in quantitative finance (2020)
  98. Quantitative day trading from natural language using reinforcement learning (2021)
  99. Reinforcement-learning based portfolio management with augmented asset movement prediction states (2020)
  100. Investor-imitator: A framework for trading knowledge extraction (2018)
  101. DeepTrader: a deep reinforcement learning approach for risk-return balanced portfolio management with market conditions Embedding (2021)
  102. Commission fee is not enough: A hierarchical reinforced framework for portfolio management (2021)
  103. Earnhft: Efficient hierarchical reinforcement learning for high frequency trading — → 2309.12891
  104. DeepScalper: A risk-aware reinforcement learning framework to capture fleeting intraday trading opportunities (2022)
  105. Universal trading for order execution with oracle policy distillation (2021)
  106. Deep reinforcement learning for automated stock trading: An ensemble strategy (2020)
  107. Deep direct reinforcement learning for financial signal representation and trading (2016)
  108. Adaptive quantitative trading: An imitative deep reinforcement learning approach (2020)
  109. Auto-GPT for Online Decision Making: Benchmarks and Additional Opinions (2023) — link
  110. MetaGPT: Meta programming for a multi-agent collaborative framework (2023) — link
  111. AppAgent: Multimodal agents as smartphone users (2023) — link
  112. MM-REACT: Prompting ChatGPT for multimodal reasoning and action (2023) — link
  113. Chameleon: Plug-and-Play compositional reasoning with large language models (2023) — link
  114. GPT-4 Technical Report (2023) — link
  115. Chatgpt (2021) — [link](https://openai.com/ research/chatgpt)
  116. GPT-4V(ision) system card (2023) — link
  117. Pangu-Agent: A fine-tunable generalist agent with structured reasoning (2023) — link
  118. Chain-of-thought prompting elicits reasoning in large language models (2022)
  119. Why Can Large Language Models Generate Correct Chain-of-Thoughts? (2023) — link
  120. Talm: Tool augmented language models (2022) — link
  121. Learn to explain: Multimodal reasoning via thought chains for science question answering (2022)
  122. Dynamic prompt learning via policy gradient for semi-structured mathematical reasoning (2022) — link
  123. Theoremqa: A theorem-driven question answering dataset (2023) — link
  124. Scibench: Evaluating college-level scientific problem-solving abilities of large language models (2023) — link
  125. Scieval: A multi-level large language model evaluation benchmark for scientific research (2023) — link
  126. Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts (2023) — link
  127. Generate rather than retrieve: Large language models are strong context generators (2022) — link
  128. Billion-scale similarity search with gpus (2019)
  129. O ptuna: A Next-Generation Hyperparameter Optimization Framework (2019)
  130. Technical analysis of stock trends (2018)
  131. Synapse: Trajectory-as-Exemplar Prompting with Memory for Computer Control (2023)
  132. Lightgbm: A highly efficient gradient boosting decision tree (2017)

2 of 140 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.