財務法說會（Earnings Conference Call, ECC）為企業與投資人溝通財務現況與未來展望之重要管道，其語意內容常對市場情緒與股價波動產生顯著影響。然由於法說會文本結構複雜、語意層次多元，傳統自然語言處理（NLP）方法在有效提取關鍵語意與預測市場反應上仍存挑戰。為此，本研究提出一套整合語意檢索、情緒量化與自適應神經架構的預測機制，結合 Retrieval-Augmented Generation（RAG）、FinBERT 與模組化單層前饋神經網路（IOSFCR_R），以建構一個可處理事件驅動語意輸入並預測短期股價波動性的結構化系統。
本研究聚焦於探討法說會語意特徵對股價短期（3 日）與中期（7 日）波動性之可預測性，並進一步驗證模型是否具備跨公司語意泛化能力（Cross-Firm Semantic Generalization）。實驗設計上，模型僅於部分公司資料上訓練，並以未見公司（AMD、ORCL）為測試對象，進行外部驗證。模型版本中，IOSFCR_R_300 經由模組重組與正則化策略，有效控制過擬合問題，於所有評估指標（MAE、MAPE、RMSE）上皆顯著優於線性回歸（Linear Regression）與雙層神經網絡（2LNN）等傳統基準模型，展現穩定的跨樣本預測能力與語意解讀能力。
本研究成果顯示，結合語意檢索與情感向量化之深度學習架構，能有效擷取法說會語意對市場波動之隱含關係，並於變動劇烈之情境下保持預測穩定性，具備處理事件驅動金融資料之潛力。未來可進一步拓展應用至多語言、多產業與非語言特徵（如聲音、技術指標）之多模態預測任務，實現更全面且具彈性的市場行為建模系統。 

Earnings Conference Calls (ECCs) serve as a crucial medium through which firms communicate their financial status and strategic outlook to investors. The semantic content expressed by corporate executives often induces significant market sentiment and short-term stock volatility. However, the inherently complex and multi-dimensional nature of ECC transcripts presents persistent challenges for traditional natural language processing (NLP) methods in effectively extracting key insights and predicting market reactions. To address this, this study proposes an integrated prediction framework that combines semantic retrieval, sentiment quantification, and an adaptive neural architecture. Specifically, we incorporate Retrieval-Augmented Generation (RAG), FinBERT, and a modularized Input-Output Self-Organizing Fully Connected Regressor (IOSFCR_R) to forecast short-term stock volatility in response to event-driven textual data.
This research focuses on examining the semantic predictability of ECCs for both short-term (3-day) and medium-term (7-day) stock volatility, while empirically testing the model’s ability to generalize across firms (cross-firm semantic generalization). The experimental setup involves training models on a subset of firms and evaluating performance on two unseen companies (AMD and ORCL), thereby providing a robust validation of generalization capability. Among the model variants, IOSFCR_R_300—with enhanced regularization and modular recomposition—effectively mitigates overfitting and consistently outperforms benchmark models such as Linear Regression and Two-Layer Neural Networks (2LNN) across all evaluation metrics (MAE, MAPE, RMSE). These findings demonstrate the model’s superior robustness and semantic interpretability in cross-firm forecasting scenarios.
The results of this study underscore the potential of integrating semantic retrieval and sentiment-aware deep learning architectures in capturing the latent relationship between financial discourse and market behavior. The proposed IOSFCR_R model maintains high predictive stability even under volatile conditions, offering a promising solution for event-driven financial forecasting. Future research may further extend this framework to multilingual, cross-sector, and multimodal prediction tasks by incorporating prosodic and market-based indicators, thereby enabling more comprehensive and flexible modeling of investor responses.
The study employs a dataset of 1,022 ECC transcripts from publicly listed firms, with the predictive target defined as stock volatility over a 3-day and 7-day post-event window. Experimental results demonstrate that the proposed IOSFCR_R-based framework outperforms traditional statistical models and non-modular neural networks across multiple evaluation metrics, including MAE, MAPE, and RMSE. The results indicate superior accuracy, interpretability, and generalization capability across different firms and time periods.
This study provides empirical evidence that the integration of semantic-driven feature extraction with modular adaptive learning significantly enhances the modeling of event-based financial texts. Future research may extend this framework by incorporating multi-modal data sources—such as audio signals, news reports, or social media—to expand its applicability across languages, industries, and dynamic market conditions.  

Abarbanell, J. S., & Bushee, B. J. (1997). Fundamental Analysis, Future Earnings, and Stock Prices. Journal of Accounting Research, 35(1), 1. https://doi.org/10.2307/2491464
Araci, D. T. (2019). FinBERT: Financial Sentiment Analysis with Pre-trained Language Models. https://arxiv.org/pdf/1908.10063
Awotunde, J. B., Adeniyi, E. A., Ogundokun, R. O., & Ayo, F. E. (2021). Application of Big Data with Fintech in Financial Services. 107–132. https://doi.org/10.1007/978-981-33-6137-9_3
Box, G. E. P. ., Jenkins, G. M. ., Reinsel, G. C. ., & Ljung, G. M. . (2016). Time series analysis : forecasting and control. https://books.google.com/books/about/Time_Series_Analysis.html?hl=zh-TW&id=rNt5CgAAQBAJ
Cao, Y., Chen, Z., Pei, Q., Lee, N. J., Subbalakshmi, K. P., & Ndiaye, P. M. (2024). Ecc analyzer: Extract trading signal from earnings conference calls using large language model for stock performance prediction. arXiv preprint arXiv:2404.18470.
Granger, C. W. J., & Teräsvirta, T. (1993). Modelling Nonlinear Economic Relationships. Modelling Nonlinear Economic Relationships. https://doi.org/10.1093/OSO/9780198773191.001.0001
Henrique, B. M., Sobreiro, V. A., & Kimura, H. (2019). Literature review: Machine learning techniques applied to financial market prediction. Expert Systems with Applications, 124, 226–251. https://doi.org/10.1016/J.ESWA.2019.01.012
Huang, A. H., Wang, H., & Yang, Y. (2023). FinBERT: A Large Language Model for Extracting Information from Financial Text*. Contemporary Accounting Research, 40(2), 806–841. https://doi.org/10.1111/1911-3846.12832
Kumar, D., Sarangi, P. K., & Verma, R. (2022). A systematic review of stock market prediction using machine learning and statistical techniques. Materials Today: Proceedings, 49, 3187–3191. https://doi.org/10.1016/J.MATPR.2020.11.399
Kumar, M., Economics, M. A.-S. in B. and, & 2014, undefined. (n.d.). An application of time series ARIMA forecasting model for predicting sugarcane production in India. Researchgate.Net. Retrieved July 3, 2025, from https://www.researchgate.net/profile/Manoj_Kumar458/publication/263505554_An_Application_Of_Time_Series_Arima_Forecasting_Model_For_Predicting_Sugarcane_Production_In_India/links/0f31753b1c8ef0afeb000000/An-Application-Of-Time-Series-Arima-Forecasting-Model-For-Predicting-Sugarcane-Production-In-India.pdf
Liu, F., Kang, Z., & Han, X. (2024). Optimizing RAG Techniques for Automotive Industry PDF Chatbots: A Case Study with Locally Deployed Ollama Models. https://arxiv.org/abs/2408.05933v1
Li, H. (2024). An advanced learning mechanism for long-term time-series forecasting (Master’s thesis, Department of Management Information Systems, National Chengchi University, Taiwan).
Mayew, W. J., & Venkatachalam, M. (2012). The power of voice: Managerial affective states and future firm performance. Journal of Finance, 67(1), 1–43. https://doi.org/10.1111/J.1540-6261.2011.01705.X;JOURNAL:JOURNAL:15406261;CTYPE:STRING:JOURNAL
Medhat, W., Hassan, A., & Korashy, H. (2014). Sentiment analysis algorithms and applications: A survey. Ain Shams Engineering Journal, 5(4), 1093–1113. https://doi.org/10.1016/J.ASEJ.2014.04.011
Nabipour, M., Nayyeri, P., Jabani, H., & Mosavi, A. (2020). Deep learning for Stock Market Prediction. Entropy 2020, Vol. 22, Page 840, 22(8), 840. https://doi.org/10.3390/e22080840
Neurocomputing, G. Z.-, & 2003, undefined. (n.d.). Time series forecasting using a hybrid ARIMA and neural network model. ElsevierGP ZhangNeurocomputing, 2003•Elsevier. Retrieved July 3, 2025, from https://www.sciencedirect.com/science/article/pii/S0925231201007020?casa_token=F2hWa1CS1UwAAAAA:aAaKrrZFPIaEKxL504c3_Vhd9EWuG--oYBKDis0UxNt1HbPLOBZ8_mr3R8iMcB6vVyBZQ0U1pA
Sohangir, S., Wang, D., Pomeranets, A., & Khoshgoftaar, T. M. (2018). Big Data: Deep Learning for financial sentiment analysis. Journal of Big Data, 5(1), 1–25. https://doi.org/10.1186/S40537-017-0111-6/TABLES/8
Tsaih, R. R. (1998). An explanation of reasoning neural networks. Mathematical and Computer Modelling, 28(2), 37–44. https://doi.org/10.1016/S0895-7177(98)00090-9
Yang, H., Zhang, B., Wang, N., Guo, C., Zhang, X., Lin, L., Wang, J., Zhou, T., Guan, M., Zhang, R., & Wang, C. D. (2024). FinRobot: An Open-Source AI Agent Platform for Financial Applications using Large Language Models. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.4841493