Paper: arXiv 2509.08742

Authors: Yanlong Wang, Jian Xu, Fei Ma, Hongkang Zhang, Hang Yu, Tiantian Gao, Yu Wang, Haochen You, Shao-Lun Huang, Danny Dongning Sun, Xiao-Ping Zhang

Abstract

Financial time series forecasting is both highly significant and challenging. Previous approaches typically standardized time series data before feeding it into forecasting models, but this encoding process inherently leads to a loss of important information. Moreover, past time series models generally require fixed numbers of variables or lookback window lengths, which further limits the scalability of time series forecasting. Besides, the interpretability and the uncertainty in forecasting remain areas requiring further research, as these factors directly impact the reliability and practical value of predictions. To address these issues, we first construct a diverse financial image-text dataset (FVLDB) and develop the Uncertainty-adjusted Group Relative Policy Optimization (UARPO) method to enable the model not only output predictions but also analyze the uncertainty of those predictions. We then proposed FinZero, a multimodal pre-trained model finetuned by UARPO to perform reasoning, prediction, and analytical understanding on the FVLDB financial time series. Extensive experiments validate that FinZero exhibits strong adaptability and scalability. After fine-tuning with UARPO, FinZero achieves an approximate 13.48% improvement in prediction accuracy over GPT-4o in the high-confidence group, demonstrating the effectiveness of reinforcement learning fine-tuning in multimodal large model, including in financial time series forecasting tasks.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper presents a novel, mathematically dense method (UARPO) with complex formulations, yet it is also highly data/implementation-heavy, featuring a custom dataset (FVLDB), specific evaluation metrics, and reported quantitative improvements over a benchmark model.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Improve Multimodal Financial TS Forecasting<br>with Uncertainty Quantification"] --> B["Key Methods:<br>1. UARPO (Reinforcement Learning)<br>2. Multimodal Pre-training"]
  B --> C{"Data/Inputs:<br>FVLDB Image-Text Financial Dataset"}
  C --> D["Computational Process:<br>1. Multimodal Pre-training<br>2. UARPO Fine-tuning"]
  D --> E["Key Findings/Outcomes:<br>1. FinZero Model<br>2. Improved Scalability & Adaptability<br>3. 13.48% Accuracy Gain vs GPT-4o"]

Cited References (43)

  1. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning (2025) — link
  2. REINFORCE++: An Efficient RLHF Algorithm with Robustness to Both Prompt and Reward Models (2025) — link
  3. Gpt-4 technical report (2023) — link
  4. Constitutional ai: Harmlessness from ai feedback (2022) — link
  5. Openai o1 system card (2024) — link
  6. Logic-rl: Unleashing llm reasoning with rule-based reinforcement learning (2025) — link
  7. Vinci. R1-v: Reinforcing super generalization ability in vision-language models with less than 3
  8. R1-Multimodal-Journey: A Journey to Real Multimodal Reinforcement Learning (2024) — link
and 35 more
  1. Lmm-r1: Empowering 3b lmms with strong reasoning abilities through two-stage rule-based rl (2025) — link
  2. Time-varying pattern causality inference in global stock markets (2021)
  3. How do crises spread? Evidence from accessible and inaccessible stock indices (2006)
  4. A Time Series is Worth 64 Words: Long-term Forecasting with Transformers (2023) — link
  5. MM-Eureka: Exploring the Frontiers of Multimodal Reasoning with Rule-based Reinforcement Learning (2025) — link
  6. The efficient market hypothesis and its critics (2003)
  7. Machine learning and the stock market (2023)
  8. Relations among Bitcoin Futures, Bitcoin Spot, Investor Attention, and Sentiment (2023)
  9. Risk, uncertainty, and expected returns (2016)
  10. Machine learning in the Chinese stock market (2022)
  11. Empirical asset pricing via machine learning (2020)
  12. Taming the factor zoo: A test of new factors (2020)
  13. Stock market prices do not follow random walks: Evidence from a simple specification test (1988)
  14. Indexing and stock market serial dependence around the world (2019)
  15. Generative Learning for Financial Time Series with Irregular and Scale-Invariant Patterns
  16. Soft Contrastive Learning for Time Series
  17. Time- LLM : Time Series Forecasting by Reprogramming Large Language Models
  18. Calf: Aligning llms for time series forecasting via cross-modal fine-tuning (2025)
  19. One fits all: Power general time series analysis by pretrained lm (2023)
  20. Are language models actually useful for time series forecasting? (2024)
  21. Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting (2021)
  22. itransformer: Inverted transformers are effective for time series forecasting (2023) — link
  23. Moderntcn: A modern pure convolution structure for general time series analysis (2024)
  24. DAM : Towards a Foundation Model for Forecasting
  25. TEMPO : Prompt-based Generative Pre-trained Transformer for Time Series Forecasting (2024)
  26. Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift (2022)
  27. Timesnet: Temporal 2d-variation modeling for general time series analysis (2022) — link
  28. Fourier GNN : Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective (2023)
  29. DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models (2024) — link
  30. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (2025) — link
  31. Proximal Policy Optimization Algorithms (2017) — link
  32. Can llms learn uncertainty on their own? expressing uncertainty effectively in a self-training manner (2024)
  33. A Survey of Uncertainty Estimation Methods on Large Language Models (2025)
  34. Can LLMs Learn Uncertainty on Their Own? Expressing Uncertainty Effectively in A Self-Training Manner (2024)
  35. Article Title (1920)

0 of 43 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2518 linked papers.