隨著生成式人工智慧技術的蓬勃發展，大型語言模型(LLM)在財務分析與企業價值評估領域展現出嶄新潛力。然而，財務預測過程在處理非結構化數據(如新聞文本)時，處理過程往往需耗費許多人力，也受限於資訊眾多而難以即時整合。本研究針對此問題，設計並實證一套LLM導入財務預測流程的新方法。本研究創新之處在於首度將LLM結合評價模型，並試算出各公司於該時點的內在價值。除此之外，本文也系統性比較不同語言模型及預測策略在企業內在價值估算上的成效差異。同時，本研究引入上下文學習機制，透過few-shot提示詞引導模型自動選擇最適合的預測方法，結果顯示，小幅(<50%)正報酬個數較原模型增加了11%。
實證部分，本研究以台灣上市櫃公司為對象，彙整2019年至2024年之歷史財報與鉅亨網新聞資料，採用gpt-3.5-turbo、gpt-4o-mini及DeepSeek-V3-0324等多款語言模型，並分別運用不同預測方式進行目標價格預測。配對t檢定結果顯示，即使在同一預測方法下，不同語言模型的預測結果在統計上亦具顯著差異。進一步比較發現，gpt-4o-mini之正報酬預測數量較gpt-3.5-turbo提升約15%，而50%以內正報酬的預測數量，則較gpt-3.5-turbo多出約88%，展現出更優異的預測能力與實務參考價值。

整體而言，本研究證實結合LLM與財務預測流程可提升對企業價值的自動化分析效率，並為傳統模型在處理非結構化資料時的侷限提供有效解方。最後，亦針對開源與商業化語言模型於財務實務應用上的優劣進行討論，並建議未來研究可結合產業專家知識以進一步優化模型成效。

With the rapid development of generative artificial intelligence, large language models (LLMs) have shown great potential in financial analysis and corporate valuation. Traditional financial forecasting often struggles to integrate unstructured data, such as news texts. This process requires significant manual effort and makes timely information synthesis difficult. This study solves these problems by designing and empirically validating a novel LLM-driven financial forecasting framework, which for the first time integrates LLMs with valuation models to compute the intrinsic value of companies. By incorporating in-context learning through prompt engineering and parameter adjustment, the proposed approach enhances model understanding of both structured and unstructured data.
Using data from listed companies in Taiwan (2019–2024), we compare gpt-3.5-turbo, gpt-4o-mini, and DeepSeek-V3-0324 across different forecasting methods. Paired t-test results reveal statistically significant differences between model outputs, even under the same forecasting strategy. Notably, gpt-4o-mini outperforms gpt-3.5-turbo, achieving about 15% more positive value predictions and 88% more predictions within the 50% positive return range.

Overall, this research demonstrates that combining LLMs with financial forecasting improves automation and overcomes limitations of traditional models in handling unstructured data. At the same time, the strengths and weaknesses of open source versus commercial models for practical financial applications are also discussed in the paper. When introducing AI models into financial analysis workflows, companies should also carefully evaluate the trade-offs between open-source and commercial language models.

Abramski, K., Citraro, S., Lombardi, L., Rossetti, G., & Stella, M. 2023. Cognitive Network Science Reveals Bias in gpt-3, gpt-3.5 turbo, and gpt-4 Mirroring Math Anxiety in High-school Students. Big Data and Cognitive Computing, 7(3), 124.
Atreides, K., & Kelley, D. J. 2024. Cognitive Biases in Natural Language: Automatically Detecting, Differentiating, and Measuring Bias in Text. Cognitive Systems Research, 88, 101304.
Bank for International Settlements, B. 2022. Market Dysfunction and Central Bank Tools. Retrieved from https://www.bis.org/publ/mc_insights.pdf
Bing, Y., Hao, J. K., & Zhang, S. C. (2012). Stock Market Prediction Using Artificial Neural Networks. Paper presented at the Advanced Engineering Forum.
Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., . . . Askell, A. 2020. Language Models are Few-shot Learners. Advances in Neural Information Processing Systems, 33, 1877-1901.
Cao, Y., Chen, Z., Pei, Q., Dimino, F., Ausiello, L., Kumar, P., . . . Ndiaye, P. M. 2024. RiskLabs: Predicting Financial Risk Using Large Language Model Based on Multi-Sources Data. arXiv preprint arXiv:2404.07452.
De Bondt, W. F., & Thaler, R. 1985. Does the Stock Market Overreact? The Journal of Finance, 40(3), 793-805.
Dolphin, R., Dursun, J., Chow, J., Blankenship, J., Adams, K., & Pike, Q. 2024. Extracting Structured Insights from Financial News: An Augmented LLM Driven Approach. arXiv preprint arXiv:2407.15788.
Fama, E. F. 1965. The Behavior of Stock-Market Prices. The Journal of Business, 38(1), 34-105.
Fama, E. F. 1970. Efficient Capital Markets. Journal of Finance, 25(2), 383-417.
Fatouros, G., Metaxas, K., Soldatos, J., & Kyriazis, D. 2024. Can Large Language Models Beat Wall Street? Unveiling the Potential of AI in Stock Selection. arXiv preprint arXiv:2401.03737.
Goldberg, A. 2024. AI in Finance: Leveraging Large Language Models for Enhanced Decision-Making and Risk Management. Social Science Journal for Advanced Research, 4(4), 33-40.
Goyal, A., & He, Z. 2015. Passive Investing and Market Liquidity. Rev Financ Stud, 28(7), 2167-2203.
Guo, B., Zhang, X., Wang, Z., Jiang, M., Nie, J., Ding, Y., . . . Wu, Y. 2023. How Close is Chatgpt to Human Experts? Comparison Corpus, Evaluation, and Detection. arXiv preprint arXiv:2301.07597.
Heston, S. L., & Sinha, N. R. 2017. News vs. Sentiment: Predicting Stock Returns from News Stories. Financial Analysts Journal, 73(3), 67-83.
Insights, C. (2025). The Foundation Model Divide: Mapping The Future of Open vs. Closed AI Development. Retrieved from https://www.cbinsights.com/research/foundation-model-divide-open-vs-closed-ai
Kim, A., Muhn, M., & Nikolaev, V. 2024. Financial Statement Analysis with Large Language Models. arXiv preprint arXiv:2407.17866.
Krause, D. 2023. Large Language Models and Generative AI in Finance: An Analysis of ChatGPT, Bard, and Bing AI. Bard, and Bing AI (July 15, 2023).
Lewellen, J. 2002. Momentum and Autocorrelation in Stock Returns. The Review of Financial Studies, 15(2), 533-564.
Noy, S., & Zhang, W. 2023. Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence. Science, 381(6654), 187-192.
Nti, I. K., Adekoya, A. F., & Weyori, B. A. 2020. A Systematic Review of Fundamental and Technical Analysis of Stock Market Predictions. Artificial Intelligence Review, 53(4), 3007-3057.
Pan, Z., Jiang, Y., Garg, S., Schneider, A., Nevmyvaka, Y., & Song, D. (2024). $ S^ 2$ IP-LLM: Semantic Space Informed Prompt Learning with LLM for Time Series Forecasting. Paper presented at the Forty-first International Conference on Machine Learning.
Sarkar, S. K., & Vafa, K. 2024. Lookahead Bias in Pretrained Language Models. Available at SSRN.
Soldatos, J., & Kyriazis, D. 2021. Trusted Artificial Intelligence in Manufacturing; Trusted Artificial Intelligence in Manufacturing: A review of the Emerging Wave of Ethical and Human Centric AI Technologies for Smart Production; A Review of the Emerging Wave of Ethical and Human Centric AI Technologies for Smart Production.
Sureshkumar, K., & Elango, N. 2011. An Efficient Approach to Forecast Indian Stock Market Price and Their Performance Analysis. International Journal of Computer Applications, 34(5), 44-49.
Tversky, A., Kahneman, D., & Slovic, P. 1982. Judgment Under Uncertainty: Heuristics and Biases: Cambridge.
Zhang, B., Yang, H., Zhou, T., Ali Babar, M., & Liu, X.-Y. (2023). Enhancing Financial Sentiment Analysis via Retrieval Augmented Large Language Models. Paper presented at the Proceedings of the fourth ACM international conference on AI in finance.
李淑蓮. (2025). 開源與閉源AI模型的戰爭：現在與未來. Retrieved from https://naipnews.naipo.com/5068#%E9%96%8B%E6%BA%90%E8%88%87%E9%96%89%E6%BA%90%E6%A8%A1%E5%9E%8B%E7%9A%84%E7%8F%BE%E7%8B%80
林美珍和張文菖. 2016. 台股目標價預測準確度之探討. 證券市場發展季刊, 28(2), 1-36.
紀麗文. (2006). 技術分析之實證研究: 檢驗 K 線戰法. 國立臺灣大學經濟學系學位論文, 1-63.
黃健峯, 林彧頎和洪宗貝. 2024. 利用生成式 AI 的金融預測: Bing Chat 在股票市場的評估與應用. 前瞻科技與管理, 12(2), 12-28.