Publications-Theses
Article View/Open
Publication Export
Google ScholarTM
NCCU Library
Citation Infomation
Related Publications in TAIR
| Title | 利用大型語言模型分析公司重大訊息並生成當沖交易策略 Analyzing Material Information with Large Language Models for Generating Day Trading Signals |
| Creator | 柯昱均 Ke, Yu-Chun |
| Contributor | 黃泓智 Huang, Hong-Chih 柯昱均 Ke, Yu-Chun |
| Key Words | 大型語言模型 生成式人工智慧 當沖交易策略 重大訊息分析 股價趨勢預測 事件研究法 機器學習 集成學習 Large language models Generative AI Day trading strategies Material information analysis Stock price prediction Event study methodology Machine learning Ensemble learning |
| Date | 2024 |
| Date Issued | 4-Feb-2025 16:03:53 (UTC+8) |
| Summary | 本研究探討生成式人工智慧（Generative AI）中的大型語言模型（LLMs）在金融市場的應用，特別是利用其分析上市公司重大訊息並生成當沖交易策略。雖然LLMs在多領域已有顯著進展，金融應用尚屬初步探索階段。本研究旨在驗證ChatGPT是否具備分析重大訊息以推斷股價趨勢的能力，並評估其交易策略效果。資料範圍聚焦於臺灣50指數成分股的重大訊息及新聞資料，並利用事件研究法進行異常報酬檢定。此外，研究結合技術指標與機器學習模型，期望透過多種訊號的整合，提升交易策略的準確性與表現。
研究結果顯示，大型語言模型在做空和多空策略中表現出顯著優勢，能有效捕捉市場中的負面訊息，尤其在市場情緒不穩定或趨勢反轉時具有較高的預測靈敏度。然而，做多策略的表現相對不佳，主要因為正面訊息的來源多樣且可能存在資訊洩漏，使得依賴重大訊息的做多策略風險較高。對於漲跌二分類任務，模型表現較為穩定，能清晰區分市場趨勢；但在漲跌三分類任務中，模型準確度有所下降，因為細緻的分類使得邊界情況下的預測更加困難。在集成學習方面，投票法雖能減少單一模型偏差帶來的風險，但也使的模型變的較為平庸。相對地，堆疊法通過結合多種機器學習模型的判斷，有效改善了做多策略的表現，並在做空及多空策略中展現優異效果。綜合來看，大型語言模型在多空策略及漲跌二分類任務中展現出應用潛力，並顯示結合歷史股價資訊的機器學習模型能夠有效提升交易決策品質。 This study explores the application of large language models (LLMs) in generative artificial intelligence (Generative AI) for financial markets, focusing on analyzing major corporate announcements and generating day trading strategies. Although LLMs have advanced significantly in various fields, their financial applications remain in the early stages. The research assesses whether ChatGPT can forecast stock price trends from material information and evaluates the effectiveness of these trading strategies. The study examines announcements and news related to Taiwan 50 Index constituents, applying event study methodology to detect abnormal returns. It also integrates technical indicators and machine learning models to enhance strategy accuracy. The results show that LLMs excel in short-selling and mixed strategies, effectively capturing negative market signals, especially during periods of instability or trend reversals. Long strategies perform less well due to varied sources of positive information and potential information leakage, increasing risk. The model is consistent in binary classification but less accurate in ternary classification due to increased complexity. While voting methods in ensemble learning reduce bias, they yield mediocre results. Stacking methods, combining multiple machine learning models, improve long strategy performance and excel in short-selling and mixed strategies. Overall, LLMs show potential in mixed strategies and binary classification, and integrating historical stock price information with machine learning models effectively enhances trading decision quality. |
| 參考文獻 | 李在僑、& 趙永祥。(2012)。現金減資宣告效果探討－以事件研究法為例。育達科大學報，(30)，103-131。 邱垂昌。(2006)。宣告及實際買回庫藏股與異常報酬－管理者之策略性應用。會計與公司治理，3(2)，17-35。 陳尚武、洪雅薰、梁嘉真、廖曉翎、劉品妤、李書瑢、& 劉涵琳。(2021)。大型海外企業併購對集團企業股價之影響－台灣鴻海併購日本夏普之實證。東亞論壇，(513)，1-13。 Akbar, M., & Baig, H. H. (2010). Reaction of stock prices to dividend announcements and market efficiency in Pakistan. Lahore Journal of Economics, 15, 103-125. Caron, M., & Müller, O. (2020, December). Hardening soft information: A transformer-based approach to forecasting stock return volatility. In 2020 IEEE International Conference on Big Data (Big Data) (pp. 4383-4391). IEEE. Cho, S. (2024, March 14). Can ChatGPT generate stock tickers to buy and sell for day trading? Available at SSRN: https://ssrn.com/abstract=4759311 or http://dx.doi.org/10.2139/ssrn.4759311 Huang, H., & Zhao, T. (2021, April). Stock market prediction by daily news via natural language processing and machine learning. In 2021 International Conference on Computer, Blockchain and Financial Development (CBFD) (pp. 190-196). IEEE. Huang, S., & Liu, S. (2019). Machine learning on stock price movement forecast: The sample of the Taiwan stock exchange. International Journal of Economics and Financial Issues, 9(2), 189. Larson, I. J. (2024). AI-nvesting: An empirical analysis with sector categorization and prompt complexity considerations assessing the predictive power of ChatGPT in stock market forecasting. CMC Senior Theses, 3491. Retrieved from https://scholarship.claremont.edu/cmc_theses/3491 Lopez-Lira, A., & Tang, Y. (2023). Can ChatGPT forecast stock price movements? Return predictability and large language models. arXiv preprint arXiv:2304.07619. Lv, D., Yuan, S., Li, M., & Xiang, Y. (2019). An empirical study of machine learning algorithms for stock daily trading strategy. Mathematical Problems in Engineering, 2019(1), 7816154. MacKinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic Literature, 35(1), 13-39. Ma, F., Lyu, Z., & Li, H. (2024). Can ChatGPT predict Chinese equity premiums? Finance Research Letters, 105631. Nti, I. K., Adekoya, A. F., & Weyori, B. A. (2020). A comprehensive evaluation of ensemble learning for stock-market prediction. Journal of Big Data, 7, 20. https://doi.org/10.1186/s40537-020-00299-5 Shaheen, I. (2006). Stock market reaction to acquisition announcements using an event study approach (Undergraduate honors thesis, Franklin & Marshall College). Franklin & Marshall College Digital Repository. Retrieved from https://digital.fandm.edu Vermaelen, T. (1981). Common stock repurchases and market signalling: An empirical study. Journal of Financial Economics, 9(2), 139-183. Xie, Q., Han, W., Lai, Y., Peng, M., & Huang, J. (2023). The wall street neophyte: A zero-shot analysis of chatgpt over multimodal stock movement prediction challenges. arXiv preprint arXiv:2304.05351. |
| Description | 碩士 國立政治大學 風險管理與保險學系 111358030 |
| 資料來源 | http://thesis.lib.nccu.edu.tw/record/#G0111358030 |
| Type | thesis |
| dc.contributor.advisor | 黃泓智 | zh_TW |
| dc.contributor.advisor | Huang, Hong-Chih | en_US |
| dc.contributor.author (Authors) | 柯昱均 | zh_TW |
| dc.contributor.author (Authors) | Ke, Yu-Chun | en_US |
| dc.creator (作者) | 柯昱均 | zh_TW |
| dc.creator (作者) | Ke, Yu-Chun | en_US |
| dc.date (日期) | 2024 | en_US |
| dc.date.accessioned | 4-Feb-2025 16:03:53 (UTC+8) | - |
| dc.date.available | 4-Feb-2025 16:03:53 (UTC+8) | - |
| dc.date.issued (上傳時間) | 4-Feb-2025 16:03:53 (UTC+8) | - |
| dc.identifier (Other Identifiers) | G0111358030 | en_US |
| dc.identifier.uri (URI) | https://nccur.lib.nccu.edu.tw/handle/140.119/155501 | - |
| dc.description (描述) | 碩士 | zh_TW |
| dc.description (描述) | 國立政治大學 | zh_TW |
| dc.description (描述) | 風險管理與保險學系 | zh_TW |
| dc.description (描述) | 111358030 | zh_TW |
| dc.description.abstract (摘要) | 本研究探討生成式人工智慧（Generative AI）中的大型語言模型（LLMs）在金融市場的應用，特別是利用其分析上市公司重大訊息並生成當沖交易策略。雖然LLMs在多領域已有顯著進展，金融應用尚屬初步探索階段。本研究旨在驗證ChatGPT是否具備分析重大訊息以推斷股價趨勢的能力，並評估其交易策略效果。資料範圍聚焦於臺灣50指數成分股的重大訊息及新聞資料，並利用事件研究法進行異常報酬檢定。此外，研究結合技術指標與機器學習模型，期望透過多種訊號的整合，提升交易策略的準確性與表現。 研究結果顯示，大型語言模型在做空和多空策略中表現出顯著優勢，能有效捕捉市場中的負面訊息，尤其在市場情緒不穩定或趨勢反轉時具有較高的預測靈敏度。然而，做多策略的表現相對不佳，主要因為正面訊息的來源多樣且可能存在資訊洩漏，使得依賴重大訊息的做多策略風險較高。對於漲跌二分類任務，模型表現較為穩定，能清晰區分市場趨勢；但在漲跌三分類任務中，模型準確度有所下降，因為細緻的分類使得邊界情況下的預測更加困難。在集成學習方面，投票法雖能減少單一模型偏差帶來的風險，但也使的模型變的較為平庸。相對地，堆疊法通過結合多種機器學習模型的判斷，有效改善了做多策略的表現，並在做空及多空策略中展現優異效果。綜合來看，大型語言模型在多空策略及漲跌二分類任務中展現出應用潛力，並顯示結合歷史股價資訊的機器學習模型能夠有效提升交易決策品質。 | zh_TW |
| dc.description.abstract (摘要) | This study explores the application of large language models (LLMs) in generative artificial intelligence (Generative AI) for financial markets, focusing on analyzing major corporate announcements and generating day trading strategies. Although LLMs have advanced significantly in various fields, their financial applications remain in the early stages. The research assesses whether ChatGPT can forecast stock price trends from material information and evaluates the effectiveness of these trading strategies. The study examines announcements and news related to Taiwan 50 Index constituents, applying event study methodology to detect abnormal returns. It also integrates technical indicators and machine learning models to enhance strategy accuracy. The results show that LLMs excel in short-selling and mixed strategies, effectively capturing negative market signals, especially during periods of instability or trend reversals. Long strategies perform less well due to varied sources of positive information and potential information leakage, increasing risk. The model is consistent in binary classification but less accurate in ternary classification due to increased complexity. While voting methods in ensemble learning reduce bias, they yield mediocre results. Stacking methods, combining multiple machine learning models, improve long strategy performance and excel in short-selling and mixed strategies. Overall, LLMs show potential in mixed strategies and binary classification, and integrating historical stock price information with machine learning models effectively enhances trading decision quality. | en_US |
| dc.description.tableofcontents | 第一章 緒論 1 第一節 研究動機 1 第二節 研究目的 2 第三節 研究流程 3 第二章 文獻回顧 4 第一節 公司重大訊息發布對股價影響文獻回顧 4 第二節 利用情感分析預測股價漲跌趨勢文獻回顧 5 第三節 機器學習與股票漲跌預測文獻回顧 8 第三章 研究方法 10 第一節 研究架構 10 第二節 重大訊息分析 12 第三節 大型語言模型 15 第四節 機器學習模型 17 第五節 集成學習 21 第六節 交易策略與資產配置 23 第七節 衡量指標說明 26 第四章 實證結果 29 第一節 事件研究法分析結果 29 第二節 大型語言模型回測結果 32 第三節 機器學習模型回測結果 43 第四節 集成模型回測結果 45 第五章 結論與建議 61 參考文獻 63 | zh_TW |
| dc.format.extent | 5381876 bytes | - |
| dc.format.mimetype | application/pdf | - |
| dc.source.uri (資料來源) | http://thesis.lib.nccu.edu.tw/record/#G0111358030 | en_US |
| dc.subject (關鍵詞) | 大型語言模型 | zh_TW |
| dc.subject (關鍵詞) | 生成式人工智慧 | zh_TW |
| dc.subject (關鍵詞) | 當沖交易策略 | zh_TW |
| dc.subject (關鍵詞) | 重大訊息分析 | zh_TW |
| dc.subject (關鍵詞) | 股價趨勢預測 | zh_TW |
| dc.subject (關鍵詞) | 事件研究法 | zh_TW |
| dc.subject (關鍵詞) | 機器學習 | zh_TW |
| dc.subject (關鍵詞) | 集成學習 | zh_TW |
| dc.subject (關鍵詞) | Large language models | en_US |
| dc.subject (關鍵詞) | Generative AI | en_US |
| dc.subject (關鍵詞) | Day trading strategies | en_US |
| dc.subject (關鍵詞) | Material information analysis | en_US |
| dc.subject (關鍵詞) | Stock price prediction | en_US |
| dc.subject (關鍵詞) | Event study methodology | en_US |
| dc.subject (關鍵詞) | Machine learning | en_US |
| dc.subject (關鍵詞) | Ensemble learning | en_US |
| dc.title (題名) | 利用大型語言模型分析公司重大訊息並生成當沖交易策略 | zh_TW |
| dc.title (題名) | Analyzing Material Information with Large Language Models for Generating Day Trading Signals | en_US |
| dc.type (資料類型) | thesis | en_US |
| dc.relation.reference (參考文獻) | 李在僑、& 趙永祥。(2012)。現金減資宣告效果探討－以事件研究法為例。育達科大學報，(30)，103-131。 邱垂昌。(2006)。宣告及實際買回庫藏股與異常報酬－管理者之策略性應用。會計與公司治理，3(2)，17-35。 陳尚武、洪雅薰、梁嘉真、廖曉翎、劉品妤、李書瑢、& 劉涵琳。(2021)。大型海外企業併購對集團企業股價之影響－台灣鴻海併購日本夏普之實證。東亞論壇，(513)，1-13。 Akbar, M., & Baig, H. H. (2010). Reaction of stock prices to dividend announcements and market efficiency in Pakistan. Lahore Journal of Economics, 15, 103-125. Caron, M., & Müller, O. (2020, December). Hardening soft information: A transformer-based approach to forecasting stock return volatility. In 2020 IEEE International Conference on Big Data (Big Data) (pp. 4383-4391). IEEE. Cho, S. (2024, March 14). Can ChatGPT generate stock tickers to buy and sell for day trading? Available at SSRN: https://ssrn.com/abstract=4759311 or http://dx.doi.org/10.2139/ssrn.4759311 Huang, H., & Zhao, T. (2021, April). Stock market prediction by daily news via natural language processing and machine learning. In 2021 International Conference on Computer, Blockchain and Financial Development (CBFD) (pp. 190-196). IEEE. Huang, S., & Liu, S. (2019). Machine learning on stock price movement forecast: The sample of the Taiwan stock exchange. International Journal of Economics and Financial Issues, 9(2), 189. Larson, I. J. (2024). AI-nvesting: An empirical analysis with sector categorization and prompt complexity considerations assessing the predictive power of ChatGPT in stock market forecasting. CMC Senior Theses, 3491. Retrieved from https://scholarship.claremont.edu/cmc_theses/3491 Lopez-Lira, A., & Tang, Y. (2023). Can ChatGPT forecast stock price movements? Return predictability and large language models. arXiv preprint arXiv:2304.07619. Lv, D., Yuan, S., Li, M., & Xiang, Y. (2019). An empirical study of machine learning algorithms for stock daily trading strategy. Mathematical Problems in Engineering, 2019(1), 7816154. MacKinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic Literature, 35(1), 13-39. Ma, F., Lyu, Z., & Li, H. (2024). Can ChatGPT predict Chinese equity premiums? Finance Research Letters, 105631. Nti, I. K., Adekoya, A. F., & Weyori, B. A. (2020). A comprehensive evaluation of ensemble learning for stock-market prediction. Journal of Big Data, 7, 20. https://doi.org/10.1186/s40537-020-00299-5 Shaheen, I. (2006). Stock market reaction to acquisition announcements using an event study approach (Undergraduate honors thesis, Franklin & Marshall College). Franklin & Marshall College Digital Repository. Retrieved from https://digital.fandm.edu Vermaelen, T. (1981). Common stock repurchases and market signalling: An empirical study. Journal of Financial Economics, 9(2), 139-183. Xie, Q., Han, W., Lai, Y., Peng, M., & Huang, J. (2023). The wall street neophyte: A zero-shot analysis of chatgpt over multimodal stock movement prediction challenges. arXiv preprint arXiv:2304.05351. | zh_TW |