Similar articles

Abstract: narrative economics suggests that financial markets are strongly influenced by evolving narratives, creating opportunities for forecasting emerging events and their economic impacts. However, existing large language model (LLM)-based approaches are inadequate in terms of systematic task decomposition and alignment with financial applications. We propose MENTOR, a multi-agent framework for event and narrative trend prediction that integrates teacher–student iterative reasoning with progressive subtasks: detecting and ranking trending events, forecasting future events from current narratives, and predicting industry index performance influenced by these events. Experiments on our self-constructed Chinese key opinion leader (KOL) articles dataset and English financial news dataset show that MENTOR consistently outperforms recent baselines such as the stakeholder-enhanced future event prediction (StkFEP) and summarize–explain–predict (SEP) frameworks in both event prediction and industry ranking tasks. In addition, the backtest results at the portfolio level show that improved event and industry forecasts can bring about a practical improvement in investment performance. These results demonstrate that incorporating structured reasoning and multi-agent feedback enables more reliable event forecasting and strengthens the connection between narrative dynamics and financial market outcomes.

MENTOR：一种基于优化推理的事件与叙事趋势预测多智能体框架

[1]Ariyo AA, Adewumi AO, Ayo CK, 2014. Stock price prediction using the ARIMA model. Proc 16^th Int Conf on Computer Modelling and Simulation, p.106-112.

[2]Bao WZD, Cao YT, Yang Y, et al., 2025. Data-driven stock forecasting models based on neural networks: a review. Inform Fus, 113:102616.

[3]Chen JL, Xiao ST, Zhang PT, et al., 2024. BGE M3-embedding: multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation. https://arxiv.org/abs/2402.03216

[4]Eimann REA, 2008. Network Event Detection with Entropy Measures. PhD Dissertation, The University of Auckland, Auckland, New Zealand.

[5]Enevoldsen KC, Chung I, Kerboua I, et al., 2025. MMTEB: massive multilingual text embedding benchmark. Proc 13^th Int Conf on Learning Representations, p.1-57.

[6]Gao YF, Xiong Y, Gao XY, et al., 2024. Retrieval-augmented generation for large language models: a survey. https://arxiv.org/abs/2312.10997

[7]Guan Y, Peng H, Wang XZ, et al., 2024. OpenEP: open-ended future event prediction. https://arxiv.org/abs/2408.06578

[8]Koa KJL, Ma YS, Ng R, et al., 2024. Learning to generate explainable stock predictions using self-reflective large language models. Proc ACM Web Conf, p.4304-4315. http://doi.org/10.1145/3589334.3645611

[9]Li ML, Li S, Wang ZHL, et al., 2021. The future is not one-dimensional: complex event schema induction by graph modeling for event prediction. Proc Conf on Empirical Methods in Natural Language Processing, p.5203-5215.

[10]Li QZ, Chao Y, Li D, et al., 2023. Event detection from social media stream: methods, datasets and opportunities. Proc IEEE Int Conf on Big Data, p.3509-3516.

[11]Liang QQ, Zhu MY, Zheng XL, et al., 2021. An adaptive news-driven method for CVaR-sensitive online portfolio selection in non-stationary financial markets. Proc 13^th Int Joint Conf on Artificial Intelligence, p.2708-2715.

[12]Liu MP, Zhu ML, Wang XY, et al., 2024. ECHO-GL: earnings calls-driven heterogeneous graph learning for stock movement prediction. Proc 38^th AAAI Conf on Artificial Intelligence, p.13972-13980.

[13]Liu NF, Lin K, Hewitt J, et al., 2024. Lost in the middle: how language models use long contexts. Trans Assoc Comput Ling, 12:157-173.

[14]Liu XY, Yang HY, Gao JC, et al., 2021. FinRL: deep reinforcement learning framework to automate trading in quantitative finance. Proc 2^nd ACM Int Conf on AI in Finance, Article 1.

[15]Llama Team, 2024. The LLaMA 3 herd of models. https://arxiv.org/abs/2407.21783

[16]Ma QC, Luo XF, Luo Y, 2014. Information entropy based the stability measure of user behaviour network in microblog. Proc 10^th Int Conf on Semantics, Knowledge and Grids, p.67-74.

[17]Mroua M, Lamine A, 2023. Financial time series prediction under Covid-19 pandemic crisis with long short-term memory (LSTM) network. Human Soc Sci Commun, 10:530.

[18]Packer C, Fang V, Patil SG, et al., 2024. MemGPT: towards LLMs as operating systems. https://arxiv.org/abs/2310.08560

[19]Qian H, Zhou HT, Zhao Q, et al., 2024. MDGNN: multi-relational dynamic graph neural network for comprehensive and dynamic stock investment prediction. Proc 38^th AAAI Conf on Artificial Intelligence, p.14642-14650.

[20]Ratner N, Levine Y, Belinkov Y, et al., 2023. Parallel context windows for large language models. Proc 61^st Annual Meeting of the Association for Computational Linguistics, p.6383-6402.

[21]Schubert E, Sander J, Ester M, et al., 2017. DBSCAN revisited, revisited: why and how you should (still) use DBSCAN. ACM Trans Datab Syst, 42(3):19.

[22]Shi Y, Wang YN, Qu Y, et al., 2024. Integrated GCN-LSTM stock prices movement prediction based on knowledge-incorporated graphs construction. Int J Mach Learn Cybern, 15(1):161-176.

[23]Shiller RJ, 2019. Narrative Economics: How Stories Go Viral & Drive Major Economic Events. Princeton University Press, Princeton, USA.

[24]Shumway RH, Stoffer DS, 2019. ARIMA models. In: Time Series (1^st Ed.). Chapman and Hall/CRC.

[25]Wang XD, Salmani M, Omidi P, et al., 2024. Beyond the limits: a survey of techniques to extend the context length in large language models. Proc 33^rd Int Joint Conf on Artificial Intelligence, p.8299-8307.

[26]Xiao GX, Tian YD, Chen BD, et al., 2024. Efficient streaming language models with attention sinks. Proc 12^th Int Conf on Learning Representations, p.1-13.

[27]Yang HY, Liu XY, Wang CD, 2023. FinGPT: open-source financial large language models. https://arxiv.org/abs/2306.06031

[28]Yang MY, Zheng XL, Liang QQ, et al., 2022. A smart trader for portfolio management based on normalizing flows. Proc 31^st Int Joint Conf on Artificial Intelligence, p.4014-4021.

[29]Ye CC, Hu ZN, Deng YH, et al., 2024. MIRAI: evaluating LLM agents for event forecasting. https://arxiv.org/abs/2407.01231

[30]Yu YY, Li HH, Chen Z, et al., 2023. FinMEM: a performance-enhanced LLM trading agent with layered memory and character design. Proc 3^rd AAAI Spring Symp Series, p.595-597.

[31]Yu YY, Yao ZY, Li HH, et al., 2024. FinCon: a synthesized LLM multi-agent system with conceptual verbal reinforcement for enhanced financial decision making. Proc 38^th Int Conf on Neural Information Processing Systems, Article 4354.

[32]Yuksekgonul M, Bianchi F, Boen J, et al., 2024. TextGrad: automatic “differentiation” via text. https://arxiv.org/abs/2406.07496

[33]Zhang WT, Zhao LX, Xia HC, et al., 2024. FinAgent: a multimodal foundation agent for financial trading: tool-augmented, diversified, and generalist. https://arxiv.org/abs/2402.18485v1

[34]Zhang YZ, Li M, Long D, et al., 2025. Qwen3 embedding: advancing text embedding and reranking through foundation models. https://arxiv.org/abs/2506.05176

[35]Zhao L, 2022. Event prediction in the big data era: a systematic survey. ACM Comput Surv, 54(5):94.

[36]Zheng LM, Chiang WL, Sheng Y, et al., 2023. Judging LLM-as-a-judge with MT-Bench and chatbot Arena. Proc 37^th Int Conf on Neural Information Processing Systems, Article 2020.

[37]Zhu FQ, Gao J, Yu CL, et al., 2023. A generative approach for script event prediction via contrastive fine-tuning. Proc 37^th AAAI Conf on Artificial Intelligence, p.14056-14064.

[38]Zhu P, Li YT, Hu YF, et al., 2024. MCI-GRU: stock prediction model based on multi-head cross-attention and improved GRU. Neurocomputing, 638:130168.