Paper: arXiv 2505.15155

Authors: Yuante Li, Xu Yang, Xiao Yang, Minrui Xu, Xisen Wang, Weiqing Liu, Jiang Bian

Abstract

Financial markets pose fundamental challenges for asset return prediction due to their high dimensionality, non-stationarity, and persistent volatility. Despite advances in large language models and multi-agent systems, current quantitative research pipelines suffer from limited automation, weak interpretability, and fragmented coordination across key components such as factor mining and model innovation. In this paper, we propose R&D-Agent for Quantitative Finance, in short RD-Agent(Q), the first data-centric multi-agent framework designed to automate the full-stack research and development of quantitative strategies via coordinated factor-model co-optimization. RD-Agent(Q) decomposes the quant process into two iterative stages: a Research stage that dynamically sets goal-aligned prompts, formulates hypotheses based on domain priors, and maps them to concrete tasks, and a Development stage that employs a code-generation agent, Co-STEER, to implement task-specific code, which is then executed in real-market backtests. The two stages are connected through a feedback stage that thoroughly evaluates experimental outcomes and informs subsequent iterations, with a multi-armed bandit scheduler for adaptive direction selection. Empirically, RD-Agent(Q) achieves up to 2X higher annualized returns than classical factor libraries using 70% fewer factors, and outperforms state-of-the-art deep time-series models on real markets. Its joint factor-model optimization delivers a strong balance between predictive accuracy and strategy robustness. Our code is available at: https://github.com/microsoft/RD-Agent.

Complexity vs Empirical Score

  • Math Complexity: 4.5/10
  • Empirical Rigor: 8.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper focuses on a multi-agent framework with architectural details and experimental results (real-market backtests, specific metrics like annualized returns), indicating strong empirical rigor, while the math is largely algorithmic and systems-oriented rather than dense theoretical derivations.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Automate Quant Strategy R&D"] --> B["Data Inputs<br>Financial Markets & Historical Data"]
  B --> C{"Research Stage"}
  C --> D["Dynamic Prompting &<br>Hypothesis Formulation"]
  D --> E["Task Mapping<br>to Domain Prior"]
  E --> F{"Development Stage"}
  F --> G["Co-STEER Agent<br>Code Generation"]
  G --> H["Real-Market<br>Backtest Execution"]
  H --> I["Evaluation & Feedback"]
  I --> J["Multi-Armed Bandit<br>Adaptive Scheduler"]
  J -->|"Iterate"| C
  I --> K["Key Findings<br>2X Higher Returns, 70% Fewer Factors"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes R&D-Agent(Q), a data‑centric multi‑agent framework that automates full‑stack quantitative strategy research and development via coordinated factor‑model co‑optimization.

Source sentence
we propose R&D-Agent for Quantitative Finance, in short RD-Agent(Q), the first data-centric multi-agent framework designed to automate the full-stack research and development of quantitative strategies via coordinated factor-model co-optimization.

Assumes

No explicit assumptions found in the text.

Data

Backtest sample: 2024 – 2025-06 · CSI 500 · equities · daily

  • CSI 300 dataset market-prices public
    2008-01-01 to 2020-08-01 · 300 large‑cap A‑share stocks in the Chinese market
    Source sentence
    we use the CSI 300 dataset, covering 300 large-cap A-share stocks in the Chinese market. The time span is split into training (Jan 1, 2008 – Dec 31, 2014), validation (Jan 1, 2015 – Dec 31, 2016), and testing (Jan 1, 2017 – Aug 1, 2020).

Recognised sources: Kaggle Wind

Costs

Modeled Costs are part of the reported headline results.

cost under $10 total for the reported results

Source sentence
at a cost under $10, R&D-Agent(Q) achieves approximately 2× higher ARR

Finds 3

Reported performance (R&D-Agent(Q)): Sharpe 2.17 · annual return 19.8% · max drawdown 6.6%

  • R&D-Agent(Q) achieves up to 2× higher annualized returns than classical factor libraries while using 70% fewer factors. performance
    Source sentence
    RD-Agent(Q) achieves up to 2X higher annualized returns than classical factor libraries using 70% fewer factors
  • The framework outperforms state‑of‑the‑art deep time‑series models on real markets. performance
    Source sentence
    outperforms state-of-the-art deep time-series models on real markets
  • Joint factor‑model optimization delivers an excellent trade‑off between predictive accuracy and strategy robustness. performance
    Source sentence
    its alternating factor–model optimization further delivers excellent trade‑off between predictive accuracy and strategy robustness

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 21 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 13 papers in the archive (12 by independent authors): AllocFlow: End-to-End Portfolio Trading Action Generation … (used as a baseline); AlphaLogics: A Market Logic-Driven Multi-Agent System for … (used as a baseline); FactorBench: A Portfolio-Aware Benchmark for Automated … (uses its method or data); Alpha-R1: Alpha Screening with LLM Reasoning via … (uses its method or data); QuantaAlpha: An Evolutionary Framework for LLM-Driven Alpha … (uses its method or data)

Code

https://github.com/microsoft/RD-Agent — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (104)

  1. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models (2022)
  2. RD2Bench : Toward Data-Centric Automatic R & D (2024)
  3. Language Models Are Few-Shot Learners (2020)
  4. Reflexion: Language Agents with Verbal Reinforcement Learning (2023)
  5. Teaching Large Language Models to Self-Debug (2024)
  6. Self-Planning Code Generation with Large Language Model (2023) — link
  7. 101 formulaic alphas (2016)
  8. Qlib: An AI-Oriented Quantitative Investment Platform (2020) — link
and 96 more
  1. AlphaFin: Benchmarking Financial Analysis with Retrieval-Augmented Stock-Chain Framework (2024)
  2. LLMFactor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction (2024) — link
  3. FactorVAE: A Probabilistic Dynamic Factor Model Based on Variational Autoencoder (2022)
  4. From Deep Learning to LLMs: A Survey of AI in Quantitative Investment — → 2503.21422
  5. AgentBench: Evaluating LLMs as Agents (2024) — link
  6. Voyager: An Open-Ended Embodied Agent with LLMs (2023)
  7. DSPy: Compiling Declarative Language-Model Calls into Self-Improving Pipelines (2023) — link
  8. Language Agent Tree Search Unifies Reasoning, Acting, and Planning in Language Models (2024)
  9. Benchmarking LLM Agents on Consequential Real-World Tasks (2024) — link
  10. 101 Formulaic Alphas (2016) — link
  11. Common risk factors in the returns on stocks and bonds (1993)
  12. Replicating anomalies (2020)
  13. …and the cross-section of expected returns (2016)
  14. Taming the factor zoo: A test of new factors (2020)
  15. Attention Is All You Need (2017)
  16. A Time Series Is Worth 64 Words: Long-Term Forecasting with Transformers (2023)
  17. iTransformer: Inverted Transformers Are Effective for Time-Series Forecasting (2024) — link
  18. Mamba: Linear-Time Sequence Modeling with Selective State Spaces (2024)
  19. MASTER: Market-Guided Stock Transformer for Stock Price Forecasting (2024) — link
  20. Thompson Sampling for Contextual Bandits with Linear Payoffs (2013)
  21. Regret Analysis of Stochastic and Non-Stochastic Multi-Armed Bandit Problems (2012)
  22. Contextual Combinatorial Bandit on Portfolio Management (2023)
  23. Risk-Aware Multi-Armed Bandit Problem with Application to Portfolio Selection (2017)
  24. Portfolio Choices with Orthogonal Bandit Learning (2015)
  25. From Deep Learning to LLMs: A survey of AI in Quantitative Investment — → 2503.21422
  26. AlphaFin: Benchmarking Financial Analysis with Retrieval-Augmented Stock-Chain Framework (2024)
  27. Factorvae: A probabilistic dynamic factor model based on variational autoencoder for predicting cross-sectional stock returns (2022)
  28. Learning multiple stock trading patterns with temporal routing adaptor and optimal transport (2021)
  29. Master: Market-guided stock transformer for stock price forecasting (2024)
  30. Graph Attention Networks (2018)
  31. Learning Phrase Representations using RNN Encoder – Decoder for Statistical Machine Translation (2014)
  32. Attention is All you Need (2017)
  33. CatBoost: unbiased boosting with categorical features (2018)
  34. DoubleEnsemble: A new ensemble method based on sample reweighting and feature selection for financial data analysis (2020)
  35. Xgboost: A scalable tree boosting system (2016)
  36. Lightgbm: A highly efficient gradient boosting decision tree (2017)
  37. 101 Formulaic Alphas (2016) — link
  38. Automate strategy finding with llm in quant investment — → 2409.06289
  39. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers (2023)
  40. iTransformer: Inverted Transformers Are Effective for Time Series Forecasting (2023)
  41. Mamba: Linear-Time Sequence Modeling with Selective State Spaces (2024)
  42. Qlib: An ai-oriented quantitative investment platform (2020) — link
  43. A Dual-Stage Attention-Based Recurrent Neural Network for Time Series Prediction (2017)
  44. Long short-term memory (1997)
  45. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks (2019)
  46. Capital asset prices: A theory of market equilibrium under conditions of risk (1964)
  47. Common risk factors in the returns on stocks and bonds (1993)
  48. On persistence in mutual fund performance (1997)
  49. Deep generative symbolic regression with monte-carlo-tree-search (2023)
  50. Deep symbolic regression: Recovering mathematical expressions from data via risk-seeking policy gradients
  51. Kaizen Programming with Enhanced Feature Discovery: An Automated Approach to Feature Selection and Feature Discovery for Prediction Models (2020)
  52. Alphaevolve: A learning framework to discover novel alphas in quantitative investment (2021)
  53. QuantFactor REINFORCE: Mining Steady Formulaic Alpha Factors with Variance-bounded REINFORCE — → 2409.05144
  54. Deep reinforcement learning for optimizing finance portfolio management (2019)
  55. Dirichlet policies for reinforced factor portfolios (2020) — link
  56. Empirical asset pricing via machine learning (2020)
  57. Exponential smoothing: The state of the art (1985)
  58. Stock Price Prediction Using the ARIMA Model (2014)
  59. Predicting the Brazilian stock market through neural networks and adaptive exponential smoothing methods (2009)
  60. Forecasting stock market movement direction with support vector machine (2005)
  61. Decision support system for stock trading using multiple indicators decision tree (2014)
  62. Optiver Realized Volatility Prediction (2021)
  63. CLVSA: a convolutional LSTM based variational sequence-to-sequence model with attention for predicting trends of financial markets (2019)
  64. Hierarchical Multi-Scale Gaussian Transformer for Stock Movement Prediction (2020)
  65. LSR-IGRU: Stock Trend Prediction Based on Long Short-Term Relationships and Improved GRU — → 2409.08282
  66. Temporal and Heterogeneous Graph Neural Network for Financial Time Series Prediction — → 2305.08740
  67. Rest: Relational event-driven stock trend forecasting (2021)
  68. Master: Market-guided stock transformer for stock price forecasting (2024)
  69. Finbert: A pre-trained financial language representation model for financial text mining (2021)
  70. Can chatgpt forecast stock price movements? return predictability and large language models — → 2304.07619
  71. LLMFactor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction (2024)
  72. AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation
  73. Auto-gpt for online decision making: Benchmarks and additional opinions (2023) — link
  74. A multimodal foundation agent for financial trading: Tool-augmented, diversified, and generalist — → 2402.18485
  75. FinCon: A Synthesized LLM Multi-Agent System with Conceptual Verbal Reinforcement for Enhanced Financial Decision Making
  76. TradingAgents: Multi-Agents LLM Financial Trading Framework — → 2412.20138
  77. A rational route to randomness (1997)
  78. Heterogeneous beliefs and routes to chaos in a simple asset pricing model (1998)
  79. Prospect theory: An analysis of decision under risk (2013)
  80. Autoregressive conditional heteroscedasticity with estimates of the variance of United Kingdom inflation (1982)
  81. On the network topology of variance decompositions: Measuring the connectedness of financial firms (2014)
  82. The variation of certain speculative prices (1997)
  83. Dynamic trading with predictable returns and transaction costs (2013)
  84. Genetic programming: On the programming of computers by means of natural selection (complex adaptive systems) (1993)
  85. Deep direct reinforcement learning for financial signal representation and trading (2016)
  86. Time series analysis: forecasting and control (2015)
  87. Short and long term stock trend prediction using decision tree (2017)
  88. Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting (2021)
  89. Informer: Beyond efficient transformer for long sequence time-series forecasting (2021)
  90. A dynamic attributes-driven graph attention network modeling on behavioral finance for stock prediction (2023)
  91. MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework
  92. Beyond self-talk: A communication-centric survey of llm-based multi-agent systems (2025) — link
  93. AlphaForge: A Framework to Mine and Dynamically Combine Formulaic Alpha Factors — → 2406.18394
  94. Alpha 158 from Microsoft Qlib
  95. Alpha 360 from Microsoft Qlib
  96. PolyModel for Hedge Funds’ Portfolio Construction Using Machine Learning — → 2412.11019

11 of 104 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.