| 摘要: | 隨著大型語言模型（Large Language Models, LLMs）於各領域快速發展，其在金融投資領域的潛力逐漸受到關注。如何評估其選股與資產配置能力，已成為學術與實務的重要課題。過往文獻多聚焦於美股市場，針對台灣股市的實證研究仍屬稀少。本研究以 OpenAI 推出的 ChatGPT-4o 為核心，探討其在台灣市場中作為選股與投資組合建構輔助工具之可行性與投資潛能。研究期間涵蓋 2017 年至 2024 年，資料來源包含台灣上市公司之調整後股價、法人買賣超及 IFRS 財務報表等資訊，並透過提示語操作，讓 ChatGPT 自主進行股票推薦、成分股挑選與權重配置，建構出 10 檔與 20 檔股票的GPT推薦權重、最小變異數與等權重投資組合。為全面檢視 GPT 投資組合之表現，本研究採用市場模型、Fama-French 三因子模型、Treynor–Mazuy 模型與 GARCH 時間序列模型進行分析。實證結果顯示，ChatGPT 所建構之投資組合在樣本期間皆展現出與基準指數台灣50相近的報酬率且有較低 β 值，並偏好小型成長股。GARCH模型亦顯示， GPT 組合具有較低的波動持續性與合理的極端風險控制能力。整體而言，ChatGPT-4o 展現出整合多元資訊與進行投資分析的潛能，能透過提示詞建構具報酬與風險平衡之投資組合，具備成為投資決策輔助工具的可行性。;As large language models (LLMs) rapidly advance across various fields, their potential in financial investment has gradually attracted attention. Evaluating their ability to select stocks and allocate assets has become an important topic in both academic and practical domains. Existing literature mostly focuses on the U.S. stock market, while empirical research on the Taiwanese stock market remains sparse. This study utilizes ChatGPT-4o, developed by OpenAI, as the core tool to explore its feasibility and investment potential in stock selection and portfolio construction within the Taiwanese market. The study period spans from 2017 to 2024, with data including adjusted stock prices of Taiwanese listed companies, institutional net buying/selling, and IFRS financial statement information. By using prompt engineering, ChatGPT autonomously generated stock recommendations, selected constituent stocks, and determined portfolio weights to construct 10- and 20-stock portfolios with ChatGPT-recommended, minimum-variance, and equally-weighted strategies.To thoroughly evaluate the performance of ChatGPT’s portfolios, this study applies the Market Model, Fama-French three-factor model, Treynor–Mazuy model, and GARCH time-series model. The empirical results show that the ChatGPT-constructed portfolios achieved returns comparable to the benchmark index, the Taiwan 50, with lower beta coefficients and a preference for small-cap growth stocks. The GARCH model also indicates that the GPT portfolios feature lower volatility persistence and reasonable control of extreme risks.Overall, ChatGPT-4o demonstrates the potential to integrate diverse information and conduct investment analysis, enabling the construction of balanced portfolios with favorable return-risk profiles. This suggests its feasibility as an auxiliary tool for investment decision-making. |
