2026-11-19 ロイヤルメルボルン工科大学(RMIT)
<関連情報>
- https://www.rmit.edu.au/news/media-releases-and-expert-comments/2026/nov/ai-stock-picks
- https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5651130
シリコン・アナリスト:AIエージェントは投資情報の格差を解消できるか?株式推奨に関する実証的分析(Silicon Analyst: Can AI Agents Level the Playing Field? Evidence from Stock Recommendations)
Yongxin Xu, Yuhao Xuan, Gaoping Zheng
Monash University、Sun Yat-sen University(Lingnan University College)、RMIT University SSRN Working Paper(96 pages)DOI:10.2139/ssrn.5651130
This study empirically examines whether generative AI agents can provide reliable stock recommendations by benchmarking three generations of ChatGPT against sell-side financial analysts. The researchers evaluated daily stock recommendations, information freshness, and the use of valuation analysis, while also examining four Gemini models. The results indicate that AI agents offer accessible recommendations but deliver inferior analytical quality compared with professional analysts. Approximately one-third of ChatGPT recommendations relied on training data rather than live searches, despite explicit instructions to search for current information. Even among recommendations based on live searches, the newest cited source was more than 100 days old in 26%–61% of cases. Only 7%–17% of AI recommendations mentioned valuation, compared with 80% of analysts’ buy recommendations. Further search instructions did not eliminate these shortcomings, and the four Gemini models performed even worse. The study also found that AI-recommended stocks attracted greater investor attention but subsequently underperformed, while the recommended companies exhibited weaker earnings. These findings highlight the gap between the accessibility of AI-generated investment advice and its analytical reliability, emphasizing the importance of information freshness, valuation analysis, and rigorous validation when applying AI to financial decision-making.


