了解股價變動的影響因子，一直是很熱門的研究議題。隨著許多文字探勘技術應用在財務領域，研究方向逐漸由過去探討公司基本面或財務面的數值資料，擴大到以文字為主的文本資料。其中，在眾多影響股價的文字資料來源中 (例如:法說會逐字稿、新聞媒體等) ，本 文著重在分析師報告的文字內容。分析師報告通常會針對公司未來發展性及長期展望進行分析並給予投資人買賣建議。然而，有時股價仍會因個別公司短期事件 (例如:即將公布公司財報、管理層異動等) 做出較大反應，因此先前 Birru and Stulz (2019) 將分析師在報告中所提到的短期內對股票有明確交易方向稱為 Trade Idea (交易想法)，並衡量這些短期交易想法的績效表現，近一步證實分析師除了選股能力之 外還具有擇時的能力。然而，文本資料的處理相對於數值資料是複雜且費時的，因此本文採用近年推出的自然語言模型 Chat GPT 來協助 文本分類工作以提高效率並且完整掌握超額報酬，此外還透過微調 (Fine-tuning) 的方式來提高模型分類準確度，最後衡量其績效是否能超越人工分類，結果證實經由自然語言模型分類出的交易想法仍然可以獲取顯著超額報酬，但其績效並無顯著超越人工分類。

Understanding the factors that influence stock price movements has always been a popular research topic. With the application of various text mining techniques in the financial sector, research has gradually expanded from examining numerical data related to company fundamentals or financial metrics to focusing on text-based data. Among the many sources of text data that affect stock prices (e.g., earnings call transcripts, news media, etc.), this paper concentrates on the textual content of analyst reports. Birru and Stulz (2019) previously referred to the short-term specific trade directions mentioned by analysts in their reports as "Trade Ideas," and evaluated the performance of these short-term trade ideas to further explore whether analysts possess timing abilities in addition to stock selection skills. Handling textual data is relatively complex and time-consuming compared to numerical data. Therefore, this paper employs the recently introduced natural language model, Chat GPT, to assist in text classification tasks, aiming to improve efficiency and comprehensively capture excess returns. Additionally, fine-tuning methods are used to enhance the model's classification accuracy. Ultimately, the performance of these classifications is evaluated to determine whether they can outperform manual classifications. The results confirm that trade ideas classified by the language model still achieve significant excess returns, but its performance does not surpass that of manual classifications.

1. Abdel-Khalik, A. R., & Ajinkya, B. B. (1982). Returns to informational advantages: The case of analysts' forecast revisions. Accounting Review, 661-680.
2. Altınkılıç, O., & Hansen, R. S. (2009). On the information role of stock recommendation revisions. Journal of Accounting and Economics, 48(1), 17-36.
3. Asquith, P., Mikhail, M. B., & Au, A. S. (2005). Information content
of equity analyst reports. Journal of financial economics, 75(2), 245-282.
4. Birru, J., Gokkaya, S., Liu, X., & Stulz, R. M. (2019). Are Analyst Short‐Term Trade Ideas Valuable?. The Journal of Finance, 77(3), 1829-1875.
5. Crane, A., & Crotty, K. (2020). How skilled are security analysts?. The Journal of Finance, 75(3), 1629-1675.
6. Daniel, K., Grinblatt, M., Titman, S., & Wermers, R. (1997). Measuring mutual fund performance with characteristic‐based benchmarks. The Journal of Finance, 52(3), 1035-1058.
7. Da, Z., & Warachka, M. (2011). The disparity between long-term and short-term forecasted earnings growth. Journal of Financial Economics, 100(2), 424-442.
8. Huang, A. H., Zang, A. Y., & Zheng, R. (2014). Evidence on the information content of text in analyst reports. The Accounting Review, 89(6), 2151-2180.
9. Jegadeesh, N., Kim, J., Krische, S. D., & Lee, C. M. (2004). Analyzing the analysts: When do recommendations add value?. The Journal of Finance, 59(3), 1083-1124.
10. Loh, R. K., & Stulz, R. M. (2011). When are analyst recommendation changes influential?. The review of financial studies, 24(2), 593-627.
11. Price, S. M., Doran, J. S., Peterson, D. R., & Bliss, B. A. (2012). Earnings conference calls and stock returns: The incremental informativeness of textual tone. Journal of Banking & Finance, 36(4), 992-1011.